Elon Web Services

And I think we're all pretty AI pilled.

我觉得我们几个都挺 AI 信徒的。

And if you're AI pilled, that means we got to build a lot more compute than the world thinks.

而如果你信了 AI 那套,那就意味着我们得建的算力,要远超这个世界所以为的。

And that these models are going to be a lot more valuable than people think.

而且这些模型的价值,也会远比人们想的要大。

You combine that with their core business, I don't know another entrepreneur or another business that's a better bet on the future, right, than SpaceX.

你再把这些跟他们的核心业务放到一起看,我想不出还有哪个创业者、哪家公司,比 SpaceX 更值得押注未来。

And so I think for most institutional investors, it's a must buy, a must own, a set it and forget it, right, in order to have a a real bet on both the space and the AI future.

所以我觉得对大多数机构投资者来说,这是必买、必须持有、买了就放着不用管的标的,这样才算真正押注了太空和 AI 的未来。

All right, here we go.

好,我们开始吧。

Early morning Silicon Valley, BG2 is back.

硅谷的清晨,BG2 又回来了。

We're chopping it up on all things tech and markets.

我们要把科技和市场的方方面面都聊个透。

To do that, I have none other than GB in the house, Gavin Baker from a Treaties.

为此,我请到了不是别人,正是 GB 本人,来自 Atreides 的 Gavin Baker。

He's brought his main guy, Andrew Fox.

他还带来了他的得力干将 Andrew Fox。

And of course, I had to draft Clark Tang into the mix, my partner, um to talk to to talk about some of the big questions of the day.

当然,我也把 Clark Tang 拉进了阵容,我的搭档,来聊聊当下的一些大问题。

You know, how should we be thinking about the SpaceX IPO?

比如,我们该怎么看待 SpaceX 的 IPO?

You know, what are the big levers?

那些关键的杠杆是什么?

There are big numbers out there for what's going to happen over the course of the next few years.

关于未来几年会发生什么,外面有一些很大的数字。

So, let's break that down a bit, help simplify it for folks.

所以我们来把它拆解一下,帮大家理清楚。

I Mythos launched yesterday.

Mythos 昨天发布了。

I want to talk a little bit about like who's up, who's down in the race for superintelligence.

我想聊聊在这场超级智能的竞赛里,谁在上升、谁在下滑。

Where are we?

我们现在到哪一步了?

What did we learn with the Mythos launch?

从 Mythos 的发布里,我们学到了什么?

And Clark was in uh in in in Taiwan last week um with Jensen at Computex and GTC.

上周 Clark 在台湾,跟 Jensen 一起参加了 Computex 和 GTC。

So, what was our takeaway there?

那我们从那儿带回了什么心得?

What's going on with GPUs, memory?

GPU、内存现在是什么情况?

Where are the bottlenecks?

瓶颈都在哪儿?

And where do we go from here?

我们接下来往哪儿走?

To start everything off, um you know, maybe just kick it over to you, Gavin, talking about the SpaceX IPO.

作为开场,先把话头交给你吧,Gavin,聊聊 SpaceX 的 IPO。

The IPO is in 2 days.

IPO 就在两天后。

Uh you're a big shareholder.

你是大股东。

Congratulations.

恭喜。

We're also a shareholder.

我们也是股东。

Uh you know, we also we expect to be buying in the IPO.

我们也打算在 IPO 里买入。

The Wall Street Journal's reporting um you know, the Goldman Sachs are both saying 160 billion in revenue in 2028.

《华尔街日报》报道,还有高盛,都说 2028 年营收会到 1600 亿美元。

Um we know that the IPO is $135 a share, 1.77 trillion.

我们知道 IPO 定价是每股 135 美元,总估值 1.77 万亿。

Um so when we think about kind of what the big levers are, there's so many moving parts in this IPO.

所以当我们去想那些关键杠杆是什么时,这次 IPO 里有太多变动的部分了。

Um nobody's better than you at just breaking it down, simplifying it.

没人比你更擅长把它拆开、讲清楚。

What are the key levers that we ought to be thinking about that you're thinking about over the course of the next few years?

有哪些关键杠杆是我们该去想的、也是你在未来这几年里会去想的?

Sure.

当然。

Um so great to be here.

很高兴来到这里。

Thank you for having me.

谢谢你邀请我。

I thought we're going to call it B BGG B, but [laughter] we can stick with BG2.

我还以为我们要叫它 B-BGG-B 呢,不过[笑]我们还是叫 BG2 吧。

I've been your house.

我一直在你这儿做客。

hey.

嘿。

All subject to revision.

一切都还可以再改。

That's okay.

没关系。

That's okay.

没关系。

So I think there's two big levers or variables that I think people should focus on.

我觉得有两个大杠杆、或者说变量,是大家该关注的。

And you know, I'm not going to comment on where I think um those variables go.

而且我不会去评论我认为那些变量会走向哪儿。

But one is um you guys have this chart.

但其中一个是,你们手上有这么一张图。

Um did you did you post this on X?

你是不是把这个发到 X 上了?

I did I did before and then we also included a new addition with uh XAI's new deals as well.

我之前就算过一次,然后这回还加进了 xAI 那几笔新交易。

Yeah.

对。

So Clark, who I've known for many years, um uh made a did a great analysis here.

Clark 我认识很多年了,他这次做了一个很棒的分析。

And he shows that XAI's deal um with Google for cloud computing uh generates more operating profit per gigawatt um than Anthropic, than Meta, than Google, than OpenAI.

他算出来,xAI 跟 Google 那笔云计算的交易,每吉瓦运营利润比 Anthropic、比 Meta、比 Google、比 OpenAI 都高。

Uh their deal

他们那笔交易——

[clears throat]

(清嗓子)

actually with uh Anthropic also generates probably more operating profit than anyone but um Anthropic.

——其实跟 Anthropic 合作的那笔,产生的运营利润可能也仅次于 Anthropic 本身。

And so you know, um the your your colleague at Altimeter, Freida, also she calculated a 55% IRR

而且你看,你们 Altimeter 的同事 Freida 还算出了 55% 的 IRR。

Mhm.

嗯。

on Claude's one.

就 Claude 那笔。

You know, if you can borrow money at 6, 7, 8% and invest in something with a 55% ARR, I'm not the most sophisticated thinker, but that math maths.

你想,如果你能以 6%、7%、8% 借到钱,再投到一个 55% ARR 的东西上,我不算最精明的人,但这账算得平。

Right.

对。

And so I think the most important variable, one of the two most important, is how quickly they bring on terrestrial data centers.

所以我觉得最重要的变量,两个最重要的之一,就是他们把地面数据中心建起来的速度有多快。

We do know from Jensen that uh Elon brings data centers up faster than anyone 122 days.

我们从 Jensen 那儿确实知道,Elon 建数据中心比谁都快,122 天。

Speed is literally cost because every day you're paying electricians and plumbers.

速度真的就是成本,因为你每一天都在给电工和管道工付钱。

That's cost.

那就是成本。

And they're now monetizing them at arguably the highest rate.

而他们现在的变现速率,可以说是最高的。

And so I think, you know, everybody should run their own math on that, but that is a massive variable.

所以我觉得,每个人都该自己去算算这笔账,但这是一个巨大的变量。

Yes.

对。

Truly massive variable.

真的是个巨大的变量。

The second thing is, you know, we have a chart and it's wildly out of date now.

第二件事是,我们有一张图,现在已经严重过时了。

It's kind of freaking amazing.

这有点吓人地惊艳。

This chart is I think is this chart from 10 days ago?

这张图……我记得是 10 天前的图?

But it like the 10 or 12 days since this chart since we made this chart which shows the Pareto curves for Opus 4.7 for coding for Codex from OpenAI.

但从我们做这张图到现在,也就 10 天或 12 天,图上画的是 Opus 4.7 做编程、OpenAI 的 Codex 做编程的 Pareto 曲线。

And now we've had Opus 4.8.

而现在我们已经有了 Opus 4.8。

It was already out of date.

图当时就已经过时了。

And now we have Fable

现在我们又有了 Fable——

Totally.

完全是。

and Mythos, which is freaking wild.

——还有 Mythos,简直离谱。

In 10 days

10 天里——

like we would have had to update the chart twice.

我们大概得把这张图更新两次。

But what the Pareto curve sure shows is how much intelligence you can get for a given amount of cost.

但 Pareto 曲线真正展现的,是给定成本下你能拿到多少智能。

And I do think being all revenue will accrue to the Pareto curve.

我确实认为所有营收都会归拢到 Pareto 曲线上。

All at least kind of frontier model revenue will accrue to the Pareto curve.

至少所有前沿模型的营收都会归拢到 Pareto 曲线上。

And this is Pareto curve for code coding.

这是编程的 Pareto 曲线。

And what I think is so impressive is that you can see in the chart that Composer 2 was Pareto dominant or, you know, at the lowest level of intelligence with very little training.

让我觉得特别厉害的是,你能在图里看到,Composer 2 在 Pareto 上是占优的,或者说,在智能最低那一档、用极少的训练就做到了。

This just reflects, and I know you know Cursor well.

这恰恰反映出——我知道你很了解 Cursor。

I think you know Cursor [ __ ] a lot better than I do.

我觉得你对 Cursor 的了解比我强多了。

A vast amount better than I do.

强太多了。

[laughter] But my understanding is that Cursor and Anthropic have more tokens of proprietary coding data than anyone else.

但据我理解,Cursor 和 Anthropic 手里的私有编程数据 token,比任何人都多。

And they have more tokens of proprietary coding data than exist on the public internet.

他们私有编程数据的 token 量,比整个公开互联网上存在的还多。

And so they fed Cursor fed um used ChemK 0.25, used their own private data, did some RL, some supervised fine-tuning and they got a really good model.

于是他们喂给 Cursor——用了 Kimi K2,用了自己的私有数据,做了一些 RL、一些监督微调,就得到了一个非常好的模型。

And then they spent 3 weeks in the Colossus 2 cluster and they got a model that 12 days ago was Pareto dominant with Composer 2.5.

然后他们在 Colossus 2 集群里跑了 3 周,得到了一个模型,12 天前它靠 Composer 2.5 在 Pareto 上就是占优的。

That's on their own benchmark.

那是在他们自己的基准上。

Um Cursor bench, so maybe take it with a grain of salt.

Cursor bench,所以也许得打个折扣看。

But I think this just suggests that the Cursor data is very valuable for coding and when it is trained, you know, to Chinchilla optimal or beyond Chinchilla optimal with reinforcement learning, you know, I think it suggests that XAI and SpaceX AI has a shot of being a real player in coding.

但我觉得这恰恰说明,Cursor 的数据对编程非常有价值,而当它被训练到 Chinchilla 最优、甚至超过 Chinchilla 最优、再加上强化学习,我觉得这说明 xAI 和 SpaceX AI 有机会在编程领域成为真正的玩家。

I mean, I think one of the interesting things is, you know, we you know, the way he answered the question, right?

我觉得有意思的一点是,他回答这个问题的方式。

We didn't talk about launch.

我们没聊发射。

Right?

对吧?

We didn't talk about Starlink or communications.

我们没聊 Starlink,也没聊通信。

Those up until really 6 months ago were the business.

直到大概 6 个月前,这些还是它的主业。

Right?

对吧?

And, you know, and then we merged in x.ai and we merged in Cursor and then we announced these deals where it was very clear he was kind of building AWS right under our nose, you know, in in in in terms of this.

然后我们把 x.ai 并进来,把 Cursor 并进来,接着又宣布了那些合作,这时就很清楚了,他其实是在我们眼皮底下悄悄搭一个 AWS 出来。

But what I want to do is go to go to Fox.

但我想先把话头交给 Fox。

Give us the breakdown.

给我们拆解一下。

Three big lines of business, right?

三大业务板块,对吧?

We've got the the the communication Starlink launch business, we've got the, you know, AI compute business and then I want to come back to x.ai that you were just clicking on.

我们有通信 Starlink 发射业务,有 AI 算力业务,然后我想再回到你刚才点开的 x.ai。

But if we just go to the core business, what do we have to assume goes right in the core business both with launch and with Starlink in order to achieve the numbers that are out there?

但如果我们只看核心业务,要想达到外面那些数字,发射和 Starlink 这两块核心业务各自得假设哪些事情走对了?

Yeah, sure.

好,当然。

So, look, I think the thing that's foundational to everything is the launch business.

我是这么看的,一切的地基是发射业务。

Right?

对吧?

This is the kind of crown jewel of SpaceX.

这可以说是 SpaceX 的皇冠明珠。

Um it's something that no one else really has, notably reusability.

这是别人都没有的东西,尤其是可复用性。

And soon rapid reusability.

而且很快会是快速可复用性。

Right?

对吧?

This is, I think, what you need to believe in to get to the economics in AI that make orbital compute something that's very economically attractive.

我觉得,你得先信这一点,才能推导出让轨道算力在经济上极具吸引力的那套账。

Outside of the idea that we are in shortage for power, shortage for chips.

撇开我们正处在电力短缺、芯片短缺这个前提不谈。

Right?

对吧?

Um so, I think rapid reusability is the main thing that we're watching for and I think most people should watch for.

所以我觉得快速可复用性是我们在盯的主要指标,也是大多数人都该盯的。

Um, you know, Elon talks about it a lot, but getting these rockets to fly at a cadence that's comparable to an airline, right?

Elon 经常谈这个,就是让这些火箭飞到接近航空公司那种发射频率。

And and Gavin has used this analogy before, but um, the old rocket industry was kind of like, imagine boarding a plane, flying to California, getting off the plane, the plane explodes after.

Gavin 之前用过这个类比,老一套的火箭行业有点像,你想象登上一架飞机飞到加州,下了飞机,然后飞机就炸了。

Um, so I think what SpaceX are ultimately trying to achieve is have a Starship fly both stages, not just the booster.

所以我觉得 SpaceX 最终想实现的,是让 Starship 两级都能飞,而不只是助推器。

Um, uh, 30, 40, 50 times before you have to retrofit that ship.

在需要翻修这艘飞船之前,能飞个 30、40、50 次。

Um, and when you do that, you're amortizing the cost of the vehicle over many flights, right?

做到这一点后,你就把飞行器的成本摊到很多次飞行上了,对吧?

And that's what brings the cost down significantly.

这才是把成本大幅拉下来的原因。

But that's a really hard problem to solve.

但这是个非常难解的问题。

Extremely difficult and look, I think the company, you know, have been loud and clear, they're going to attempt to bring back the second stage

极其困难,而且我觉得公司已经说得很清楚了,他们打算把 Starship 的第二级带回来,

of Starship later this year.

就在今年晚些时候。

Um, and then make it reusable, you know, re-fly the second stage next year.

然后让它可复用,明年再把第二级重新飞一次。

Um, and from there, ramp up the cadence.

从那以后,再把发射频率往上拉。

But at the end of the day, driving down the cost of launch is what enables all of these other businesses and is what makes them so attractive relative to incumbents.

但归根结底,把发射成本压下来,才是让所有这些其他业务成立的关键,也是它们相对现有玩家如此有吸引力的原因。

So how many how many times are we Starship just launched Starship 3, you know, just launched.

那我们要发射多少次?Starship 刚发射了 Starship 3,刚刚发射的。

How many launches are you know, do you think kind of the consensus out there is assuming, you know, two or three years from now?

你觉得,比方说两三年后,外面的共识大概假设能有多少次发射?

Like what is the launch cadence?

发射频率到底是多少?

Are we launching one of these every day or we launching one of these every week or every month?

我们是每天发一枚,还是每周发一枚,还是每月发一枚?

Like where are we in terms of expectations?

就预期而言,我们现在处在什么水平?

Yeah, so look, I think expectations for now, you know, we're going from, you know, call it 160 165 launches last year up into the high hundreds of launches in several years and, you know, getting into the thousands of launches probably in the next 3 years thereafter.

好,我是这么看的,就目前的预期,我们是从去年大概 160 到 165 次发射,几年内涨到几百次的高位,再往后大概三年内,很可能进入上千次发射的量级。

Okay.

好。

Um I think the company have aspirations.

我觉得这家公司是有野心的。

Thousands of launches, you're launching, you're doing two or three launches a day.

上千次发射,你要发射,一天做两三次发射。

Right.

对。

And then talk to us a little bit, what is this enable?

那你给我们讲讲,这能带来什么?

Obviously, you know, I'm here in Silicon Valley.

很显然,我人就在硅谷。

I can't even I can't even keep a call on Sand Hill Road, two decades into the mobile revolution.

移动革命都过去二十年了,我在沙丘路上连个电话都保不住。

I mean, it's the craziest thing.

这真是最离谱的事。

It's like a third world.

简直跟第三世界一样。

It's It's a major business problem when you're freaking out here.

你在这儿信号急得抓狂,这可是个大问题。

[laughter]

(笑)

It's crazy.

太离谱了。

It's crazy.

太离谱了。

Right by the Starwood dead zone.

就在 Starwood 那片信号盲区旁边。

And I'm like, how can this possibly be?

我心想,这怎么可能?

So almost like it's a joke.

简直像个笑话。

It's the epicenter of technology in America and you can't maintain a call.

这是全美科技的震中,你却连个电话都打不通。

Okay, so we're all going to switch to Starlink mobile when it comes along because I don't want to lose that call on Sand Hill Road.

好,所以等 Starlink 手机服务出来,我们都得换,因为我不想在沙丘路上再掉线了。

So walk me through a little bit just again high-level.

那你再从高层面给我捋一捋。

Um it's a big portion of the revenue growth expected in the business over the course of the next two to three years.

未来两三年,这块占了公司预期营收增长的很大一部分。

My hunch is a lot of this is driven uh by uh by direct to cell connectivity.

我的直觉是,这里面很大一块是由直连手机的连接驱动的。

Walk me through a little bit those economics.

你给我捋一捋这里面的账。

Yeah, so look, it's actually interesting.

好,你看,这其实挺有意思。

Um the broadband business is still very early stage when you think about um the percent of households that have actually been penetrated to date.

宽带这块业务还非常早期,你想想到今天为止真正渗透进去的家庭占比。

You look at the percent of global households with Starlink, it's less than 1%.

你看全球家庭里用 Starlink 的比例,不到 1%。

Uh and that's the broadband, you know, you kind of have a base terminal at your house, on your car, on your boat, um and then airlines now as well.

这就是宽带,你在家里、车上、船上装个基础终端,现在航空公司也用上了。

Um so I actually think broadband can scale to hundreds of millions of terminals, hundreds of millions of users.

所以我其实认为宽带能扩到几亿个终端、几亿用户。

And today the subscriber base

而今天的用户基数——

Hundreds of millions if if they get rapid reusability of Starship, which is really hard.

几亿——前提是他们做到 Starship 的快速可复用性,这非常难。

Um you know, if if there's not competition.

前提是没有竞争。

Um hundreds of millions, um it's possible.

几亿,是有可能的。

But maybe

但也许——

I I always say around here, it's funny.

我在这儿老爱说一句话,挺逗的。

I love seeing PM and kind of analysts in in this situation.

我特别喜欢看 PM 和分析师碰上这种情况。

It's exactly what I do with Clark.

这正是我跟 Clark 之间的日常。

Clark will say something, I'll say the future is a distribution of unknown probabilities.

Clark 说点什么,我就说,未来是一个未知概率分布。

It's either more likely or less likely, so give me the distribution.

要么更可能,要么更不可能,那你给我这个分布。

Are we talking 20% 30%?

我们说的是 20% 还是 30%?

It's hilarious.

太逗了。

It's the same

一模一样——

Well, no, 100% same thing.

不,完全是一回事。

And like I've watched Elon do many hard things and this is a really hard thing.

我看着 Elon 做成过很多难事,而这是件真正的难事。

So, I think it's reasonable to think that they're going to succeed with rapid reusability, but just I just think it's important to acknowledge that like orbital compute you know, Starlinks, you know, Starlink V3, Starlink direct to cell, we need we need first reusability for Starship V3 and then rapid reusability unlocks a lot of this.

所以我觉得,认为他们会在快速可复用性上成功是合理的,但我只是想强调,必须承认,像轨道计算、Starlink、Starlink V3、Starlink 直连手机,我们需要先做到 Starship V3 的可复用性,然后快速可复用性会解锁这里面很多东西。

Right.

对。

When I see when I see the models that the banks are putting out there, right?

当我看到那些银行放出来的模型,对吧?

And Wall Street Journal, everybody's reported on these.

还有《华尔街日报》,大家都报道过这些。

These same things have been widely leaked.

这些同样的东西早就被广泛泄露了。

They they they largely have the revenue on connectivity, so let's call it Starlink direct to cell, etc.

他们基本上把连接这块的营收——就叫它 Starlink 直连手机之类的——

Going from, you know, let's call it 10 billion to 50 billion uh by 2028.

到 2028 年从大概 100 亿做到 500 亿。

And so, I'm not asking you guys to react to, you know, to tell me your specific numbers, but when I'm talking to Clark all I'm trying to size up is order of magnitude.

所以我不是要你们对具体数字表态,但我跟 Clark 聊的时候,我想估的只是一个数量级。

Do we think we can 5x the business over the course of the next 3 years?

我们觉得能在未来 3 年把这块业务做到 5 倍吗?

Is there enough TAM both in terms of broadband and direct to consumer?

无论是宽带还是直连消费者,TAM 够不够大?

And I think the answer to that is yes.

我觉得答案是够。

Yeah, here's what I just say very simply is I have um I I travel with Starlink.

是,我就特别简单地说,我出门都带着 Starlink。

Um I'm I'm a big video gamer and very consistently wherever I am in the world, Starlink is the best connection.

我是个重度游戏玩家,而且非常稳定,不管我在世界哪个角落,Starlink 都是最好的连接。

It's the fastest, it's lowest latency and I do think once they get to rapid reusability, it's also going to be they're going to have the cheapest cost per gigabyte or megabyte delivered um and better faster, cheaper has been a winning formula.

它最快、延迟最低,而且我确实认为,一旦他们做到快速可复用性,他们每 GB 或每 MB 的交付成本也会是最便宜的,而更好、更快、更便宜一直是制胜公式。

And so, 50 billion, that's, you know, 0.3% penetration of the global telecom market.

那么 500 亿,这也就是全球电信市场 0.3% 的渗透率。

Now, maybe there's some deflation with Starlink pricing.

当然,Starlink 定价也许会有些通缩。

Um but that's the way I'd frame it up.

但我会这么来框定它。

Yeah.

是。

I like betting on better, faster, cheaper.

我喜欢押注更好、更快、更便宜。

Um, Clark, I would say probably the biggest surprise of the last six six weeks is that Elon, you know, we talked about it on all-in podcast, we called it EWS, Elon web services, right?

Clark，我觉得过去这六周最大的意外可能就是 Elon，我们在 all-in podcast 上聊过，管它叫 EWS，Elon web services，对吧？

That that he struck these huge deals with Anthropic and Google.

就是他跟 Anthropic 和 Google 谈成的那些巨额合作。

I don't even think people were thinking about SpaceX in the AI compute game, right?

我觉得根本没人想到过 SpaceX 会进 AI 算力这个局，对吧？

We If you looked at the models as of a few months ago, it was connectivity, so Starlink, and then it was x.ai, the model.

几个月前你去看那些模型，一块是连接，也就是 Starlink，另一块是 x.ai，那个模型。

But this whole category of taking all of this compute, which he's uniquely good at standing up, right?

但把这么大规模的算力搭起来——这事他天生就特别擅长——

And then reselling it in a way that's highly profitable was not in a lot of people's forecast.

然后再用一种高利润的方式转售出去，这整个品类根本不在很多人的预测里。

Now it's a major component of the forecast.

现在它成了预测里的一个主要构成部分。

You know, you and I did this podcast with Jensen, where Jensen said Elon is an N of 1.

你和我做过那期跟 Jensen 的播客，Jensen 说 Elon 是 N of 1。

What they achieved is is singular.

他们做到的这件事是独一份的。

Never been done before.

前所未有。

Just to put in perspective, 100,000 GPUs, that's you know, easily the fastest supercomputer on the planet as one cluster.

说个直观的对比，10 万块 GPU，作为单一集群，那轻轻松松就是地球上最快的超级计算机。

Um, a supercomputer uh, that you would build would take normally 3 years to plan.

而你要建这么一台超级计算机，光规划正常就得 3 年。

And then they deliver the equipment, and it takes 1 year to get it all working.

然后他们把设备交付过来，再花 1 年才能让整套跑起来。

Yes.

对。

We're talking about 19 days.

我们说的是 19 天。

Wow.

哇。

N of 1 is right.

N of 1 说得没错。

Elon is an N of 1.

Elon 就是 N of 1。

And his ability to secure supply, stand up the supply, you know, deploy it in a way that's uh, you know, coherent and effective for both himself and I guess now for others.

还有他锁定供应、把供应搭起来、再以一种连贯又高效的方式部署下去的能力——既是为他自己，我猜现在也是为别人。

So, walk us through kind of that.

那你给我们捋一捋这块。

It looks to me again like this is a major component of the revenue story.

在我看来这又是营收故事里的一个主要构成部分。

Totally.

完全同意。

I mean, so we we were all at the macro hard data center, and it was just very evident the amount of engineering that was that had gone into building these sites. you people always talk about Google and their ability to to build a TPU and sell the TPU to Anthropic to generate revenues for AI.

我们当时都在 macro 那个硬核数据中心，一眼就能看出建这些站点投入了多少工程量。大家老爱说 Google 有能力造 TPU，再把 TPU 卖给 Anthropic，靠 AI 挣营收。

I think it's a pretty similar dynamic here with Elon able to secure power, uh build these sites faster than anyone else and also be able now to monetize it to um to the this massive AI market that's ahead of us. if you look at the the relationships that he's forged with a lot of his suppliers, you know, be it Jensen, be it, you know, all of these different um different sites that actually want xAI as a tenant.

我觉得这里的逻辑挺像的，Elon 能锁定电力、比谁都更快把这些站点建起来，现在还能把它变现给我们面前这个巨大的 AI 市场。你看他跟一大堆供应商建立的关系，不管是 Jensen，还是那些真心想让 xAI 来当租户的各个站点。

Um his his ability to finance these deals at at very attractive uh financing rates relative to a lot of the other players in in the space, you know, these are advantages that compound over time.

还有他相对场内很多其他玩家、能用非常有吸引力的融资利率给这些交易融资的能力，这些优势都会随时间复利累积。

And when you've built the credibility to stand up these sites and monetize at these levels, um you know, it's very it's actually a very attractive uh very attractive proposition for for a lot of folks involved.

而当你已经建立起把这些站点搭起来、并在这种量级上变现的信誉，那对很多相关方来说，这其实是个非常有吸引力的提议。

Um and actually, you know, if you look at the these deals in particular, Gavin, you you pointed out, but, you know, they're they're actually monetizing, you know, perhaps better than other players in the space by selling this infrastruc-

而且实际上，Gavin，你也指出过，你看这几笔交易，他们靠卖这套基础设施，变现能力其实可能比场内其他玩家还要好——

a lot higher.

高出一大截。

Um Google is obviously paying SpaceX a huge premium for this compute.

Google 显然在为这些算力给 SpaceX 付一大笔溢价。

Fox, you said something that I thought was really important, which is, you know, it may very well be that in order to get, you know, first in line on space compute, which Google certainly wants to do, that they're willing to pay a premium for their terrestrial compute.

Fox，你说过一句我觉得特别重要的话，就是很可能为了在太空算力上排到第一位——Google 当然很想这么做——他们愿意为地面算力多付一笔溢价。

And so, to me, that's how you kind of square the circle as to why the premium.

所以在我看来，这就把溢价这件事说圆了。

Any thoughts?

有什么想法吗？

Yeah, look, I think there's some of that embedded there, but um look, at the end of the day, SpaceX can stand up compute quickly.

是这样，我觉得这里面确实有一部分是这个因素，但说到底，SpaceX 能快速把算力搭起来。

They can stand it up coherently.

他们能连贯地搭起来。

And they can stand up a lot of it in one place and have it readily available.

他们还能在一个地方搭起一大批，随时可用。

So, look, I think that's most of the premium, but outside of that certainly people are going to space over time.

所以我觉得溢价大部分来自这个，但除此之外，随着时间推移人们肯定是要往太空走的。

I have to pay a little call option to get first in line for space.

我得付一点小小的看涨期权，换个进太空的排头位置。

There you go.

就是这个理。

Good one.

说得好。

We've all been investing in the neo cloud space.

我们几个都一直在投 neo cloud 这个领域。

So, like there's a fundamental belief around this table I that that we lack the compute needed to continue to push the frontier on intelligence.

所以这桌人有个根本共识，就是我们缺继续把智能推向前沿所需要的算力。

So, we have to build a lot of compute, okay?

所以我们得建一大堆算力，对吧？

Now there's a there's competition going on.

现在这里面有竞争在发生。

On one end you have the hyper scalers who are building out that capability.

一端是 hyperscaler，他们在建这个能力。

Then we have AI dedicated clouds that are building out that capability.

然后有 AI 专用云，也在建这个能力。

And now literally in a matter of weeks, right, we have a a you know a giant that's emerged in this category which is SpaceX.

现在真的就在短短几周里，这个品类里冒出来一个巨头，就是 SpaceX。

The question to you Gavin is can they consolidate this market, right?

Gavin，问你的问题是，他们能不能把这个市场整合掉？

Because if I think about a marketplace, Elon has a unique ability to get the supply.

因为如果我把它当成一个市场来想，Elon 有拿到供应的独特能力。

He has a unique ability to cut deals on the other side and nobody can stand it up like he can stand it up.

他在另一端有谈成交易的独特能力，而且没人能像他那样把它搭起来。

So, I think there might be a real consolidation in the AI compute market where you have the hyper scalers on the one hand and on the other hand, you know, he may emerge as the largest, strongest player in the AI compute market.

所以我觉得 AI 算力市场里可能会有一次真正的整合，一边是 hyperscaler，另一边，他可能会成长为 AI 算力市场里最大、最强的玩家。

Yeah, so I think they're are they the number four number five hyper scaler today after the Google deal?

对，所以我觉得，在 Google 这笔交易之后，他们现在是不是第四、第五大 hyperscaler 了？

Um it will be number four.

会是第四。

Kind of wild.

挺离谱的。

In 30 days we went from not being an AI hyper scaler to being number four.

30 天里我们从不是 AI hyperscaler 变成了第四。

And we passed a lot of companies including Oracle.

而且我们超过了一堆公司，包括 Oracle。

Coreweave is a huge business, right?

Coreweave 是个巨大的生意，对吧？

That we're we're investors in, you know, and have been investors in, right?

我们是它的投资人，而且一直都是，对吧？

But there are a lot of other players, the Nebius's of the world, the Iron's of the world.

但还有一大堆别的玩家，Nebius 那一类的，Iron 那一类的。

And I would say that they're probably 50 neo labs being funded in Silicon Valley right now as we speak because of the shortage in compute.

我要说，就在我们说话这会儿，硅谷可能有 50 家 neo lab 正在拿融资，就因为算力短缺。

Absolutely.

千真万确。

So, that's kind of crazy in 30 days.

所以 30 天里就这样，挺疯的。

That's just extraordinary.

这简直是非同寻常。

What I would say is there I think there is a belief that these data centers are commodities.

我想说的是,有一种看法认为这些数据中心是同质化的大宗商品。

And I do not share that belief.

而我不认同这种看法。

Um I don't think anybody around this table shares that belief.

我觉得在座各位没人认同这种看法。

And in the same way that Elon was able to re-engineer a rocket from first principles and make it reusable, he engineered an electric car from first principles.

就像 Elon 能从第一性原理重新设计一枚火箭、让它可复用一样,他也从第一性原理设计了一辆电动车。

You know, everyone else was trying to, you know, make an electric car like an internal combustion engine car and he thought about it differently.

别人都在照着内燃机车的思路去造电动车,而他的思考方式完全不同。

And um I think he looked at data center design from first principles and he designed something fundamentally different.

我觉得他也是从第一性原理去看数据中心的设计,设计出了一个根本不同的东西。

And I did actually ask the team.

我确实问过团队。

I said, "Hey guys, maybe I'd be a little less public about things that are very obvious to you [laughter]

我说:"各位,那些在你们看来再明显不过的事,也许我该少在公开场合讲一点,

about how to design a data center, but are revelations to other people because I think what you're doing is maybe um more differentiated than you perhaps realize cuz what you're doing is so logical to you, but maybe lot not logical to everyone else.

讲怎么设计数据中心,但这些对别人来说其实是新鲜的启示。因为我觉得你们做的东西差异化程度可能比你们自己意识到的还要大,你们觉得这一切太合乎逻辑了,但对别人来说未必。"

And that's how he was able to do it in 122 days.

他就是这样在 122 天里做成的。

Yeah, to I mean to that point, Brad, yesterday we were meeting one of our portfolio companies and we were talking about behind the meter and we're, you know, really thinking about it.

是啊,Brad,说到这点,我们昨天见了一家投资组合里的公司,聊到表后(behind-the-meter),我们真的在认真琢磨这件事。

There's only maybe two or three two or three players now that can actually reliably engineer behind the meter data center.

现在真正能可靠地设计出表后数据中心的,可能就那么两三家玩家。

And you know, there's real engineering work that goes into all of this.

这里头是有真材实料的工程活儿的。

So, if you think about this, if you're a gas combustion if you're Vernova and you say we only have a certain number of gas combustion engines.

所以你想想,如果你是做燃气发电的,如果你是 Vernova,你说我们的燃气发电机组数量就这么多。

Now, we can sell them to x.ai or we can sell them to one of these startup neo clouds.

现在,我们可以卖给 x.ai,也可以卖给这些初创的 neo cloud。

Who are you going to sell them to?

你会卖给谁?

Well, and there's another dynamic.

而且还有另一层动态。

Everyone starts making more money when the GPUs get energized and sold faster.

当 GPU 更快地通上电、卖出去,所有人都开始赚更多钱。

So, literally speed is money for all of the suppliers.

所以对所有供应商来说,速度就是真金白银。

Power, land, turbines.

电力、土地、涡轮机。

So, I think it's we'll we'll see.

所以我觉得,咱们走着瞧吧。

Hey Brad, man.

嘿 Brad,老兄。

And but but this is just we're just talking terrestrial.

但这都还只是在聊地面上的东西。

I do I do want to hit on and then you can flip it back on me.

我确实想聊一下,然后你可以再把话头抛回给我。

Talk to me, okay, So, let's let's assume, right, that they continue to build out the terrestrial landscape.

跟我说说,好,那我们就假设,他们继续把地面这块铺开建下去。

They continue to find buyers for that.

他们继续给这些产能找到买家。

Um, walk us through, you know, what this unlocks, you know, and how this is related to space data centers because I think, you know, once you start talking terrafab capacity and beyond.

那你给我们捋一捋,这解锁了什么,以及这跟太空数据中心有什么关系。因为我觉得,一旦你开始聊 terafab 级别的产能乃至更大。

So, we're talking 1,000 gigs, right?

那我们说的可是 1,000 吉瓦,对吧?

And this year what what we're doing, 25 or 30 gigs just to put it all in perspective.

而今年我们做的是多少,25 或 30 吉瓦,就给个参照。

20, yeah.

20,对。

Right?

对吧?

20, 25 gigs.

20、25 吉瓦。

Okay, so so once we start scaling up, walk us through, do we have to have space data centers in order to get excited about buying the IPO, right?

好,那一旦我们开始往上扩,你给我们捋一捋,我们是不是非得有太空数据中心,才值得为买这个 IPO 兴奋起来?

And then there's obviously this this debate in the world.

然后显然,世界上还有这么一场争论。

I I heard Jeff Bezos say, you know, I think it's more like 6 years, but Elon's going to say three because if if he says six, then it will take even longer.

我听 Jeff Bezos 说,我觉得更像是 6 年,但 Elon 会说三年,因为如果他说六年,那实际就会拖得更久。

So, say three and we may get it in four or five.

所以说三年吧,我们可能四五年能拿到。

But are space data centers integral and essential to, you know, the IPO?

但太空数据中心对这个 IPO 是不是不可或缺、至关重要?

And what do you think the timeline is, Erin?

那你觉得时间线会是怎样,Erin?

There are you guys.

各位,那边有……

So, I don't think I think if you think about those variables around what Crusher could mean for XAI.

所以如果你把 Cursor 对 xAI 意味着什么这些变量都考虑进去。

And we do have an existence proof that once you really get on that Pareto frontier, revenue can scale rapidly and it's called Entropic.

而且我们确实有一个现成的例证:一旦你真正站上那条 Pareto 前沿,营收就能快速放大,这家公司叫 Anthropic。

And there does seem to be an exhaust There seems to be a lot of demand for coding.

而且看起来确实有一大块需求存在,编程这块的需求很旺。

And I do think I'm John Massad posted something very interesting.

我确实觉得——John Massad 发了个特别有意思的东西。

The The founder of Replit.

就是 Replit 的创始人。

The founder of Replit.

Replit 的创始人。

It he called it bitter lesson adjacent that coding may be the fastest path to AGI and ASI because if you really go to coding, you can write code if a model's good at coding to do anything.

他把这叫做"接近苦涩的教训":编程可能是通往 AGI 和 ASI 最快的路径,因为如果一个模型真的很擅长编程,你就能写代码去做任何事。

So, I think that's a profound point and I think coding is going to continue to be very important.

所以我觉得这是个很深刻的观点,编程会继续非常重要。

So, I think if you think about that variable, if you think about Starlink direct to cell enabled by Starlink V3, and you think about how quickly they can or cannot bring on terrestrial compute, I I think orbital compute is is is necessary for the IPO valuation, but it's certainly important and it's

所以如果你把这个变量、把靠 Starlink V3 实现的 Starlink 直连手机、以及他们能多快建起地面算力都考虑进去,我觉得轨道算力对这个 IPO 估值来说是必要的,它当然很重要,而且它——

Well, maybe another way to say it is you think you you may think we're going to get to ASI faster than we're going to get to orbital compute.

或者换个说法,你可能觉得我们达到 ASI 会比建成轨道算力更快。

That may take us from 300 IQ to 400 IQ, 500 IQ, and beyond.

那可能把我们从 300 IQ 带到 400 IQ、500 IQ,甚至更高。

Um and the ability to scale it up to consume 10% of, you know, global GDP, but maybe maybe that's where we should move next.

还有把它放大到消耗全球 GDP 10% 的能力,不过也许我们接下来该聊这个。

No, no, I think on the orbital compute, I think Foxy would be great our Clark to lay out the math from first principles on, you know, Clark has this great chart on, you know, the gigawatts it costs, you know, the dollars per gigawatt.

不不,轨道算力这块,我觉得让 Foxy 或者 Clark 从第一性原理把这笔账讲清楚会很棒,Clark 有一张很棒的图,讲的是要花多少吉瓦、每吉瓦多少钱。

Walk us through the economic case.

给我们讲讲这里面的经济账。

Yeah.

好。

Yeah, so I mean, on this point of is orbital key to investing here?

好,那我说说——关于轨道算力是不是这里投资的关键这一点。

I don't think it is and I'll first point I'll make is what are the implied monetization rates based on expectations today for the AI business?

我不觉得是,我要说的第一点是:按今天对 AI 业务的预期,隐含的变现速率是多少?

And you know, I think you threw out the $160 billion number that's been leaked out there that people are talking about.

你之前抛出过 1600 亿美元这个数字,就是外面泄露出来、大家在讨论的那个。

The implied monetization rate on that number is something like $14 billion per gigawatt per year for the AI business.

按那个数字算,AI 业务隐含的变现速率大概是每吉瓦每年 140 亿美元。

They just signed Anthropic at 22 to 23.

他们刚跟 Anthropic 签了 22 到 23。

They just signed Google at 50.

他们刚跟 Google 签了 50。

Right.

对。

So, I I think you can invest behind the AI business terrestrially and still be excited about it.

所以我觉得你可以只押注地面上的 AI 业务,照样能为它兴奋。

But with orbital

但要是加上轨道——

an important point.

一个很重要的点。

Excited about it if they can get the land and the power.

前提是他们能拿到土地和电力,那才值得兴奋。

Right.

对。

But but but I mean I I think for most investors, right?

但我是说,对大多数投资人来说,对吧?

They get They have an easier time getting their head around how SpaceX wins terrestrially.

他们更容易想明白 SpaceX 在地面上是怎么赢的。

Like can they go get land, power, and chips?

比如他们能不能拿到土地、电力和芯片?

The answer to that is high probability yes, okay?

这个问题的答案是大概率能,对吧?

And what we're saying is at the rate they're monetizing that, that gets you to the numbers that are being leaked out there before you even have to take the leap of faith that they're going to extend the lead with orbital data centers.

我们想说的是,按他们现在的变现速率,还没等你去赌他们会用轨道数据中心把领先优势拉得更大,光这一块就已经能撑到外面泄露的那些数字了。

But take us there on that, too.

但那块你也给我们讲讲。

Sure.

好。

Yeah, so so look, with orbital, I think the key thing is um two-stage

好,那这样,轨道这块我觉得关键在于——两级。

And beyond that, rapid two-stage

再往后,是快速的两级。

So, today with Starship, they've shown that they can successfully reland the booster.

今天靠 Starship,他们已经证明能成功回收一级助推器。

The second stage, we'll see what happens later this year.

二级呢,今年晚些时候看情况。

I think they're attempting to bring that back and then make it reusable by next year. but the thing that's important about two-stage reusability when it comes to the economics for orbital compute, right, is the cost per kg comes down significantly.

我觉得他们打算把二级也带回来,然后争取明年做到可复用。但两级可复用性之所以对轨道算力的经济账重要,是因为每公斤成本会大幅下降。

You know, we're talking about going from $1,500 per kg on Falcon, somewhere in that range, to 250 per kg, something lower.

我们说的是从 Falcon 上每公斤 1500 美元、差不多那个区间,降到每公斤 250 美元,甚至更低。

Um and the more that you can reuse the rocket, the more that price comes down.

而且火箭能复用的次数越多,这个价格降得越多。

Right, cuz you're just depreciating the cost of the launch.

对,因为你只是在摊销这次发射的成本。

And eventually, you asymptote to the cost of the fuel.

最终你会渐近逼近燃料的成本。

Assuming you can use a rocket for forever.

前提是你能永远用同一枚火箭。

Right, which will take a very long time for us to to really achieve that.

对,而要真正做到这一点,我们还得花很长时间。

But, um and at that point, we're talking about something well south of 250 per kg. then you look at the specs of these AI satellites.

但到那时候,我们说的就是远低于每公斤 250 美元了。然后你再看看这些 AI 卫星的规格。

You know, Elon did a great

你知道,Elon 做了个很棒的——

Yeah, that pod that that pod was incredible that he laid out the other day, the specs on the satellites.

对,他前几天摆出来的那个节目太精彩了,讲卫星的规格。

It was really great because I think they are finally showing people, here's how you could viably design one of these satellites.

真的特别好,因为我觉得他们终于开始让大家看到:这些卫星可以怎么切实可行地设计出来。

And how heavy is the satellite?

那卫星有多重?

How many could you fit into a Starship launch?

一次 Starship 发射能塞进去多少颗?

And when you back into the numbers, you get to something like 5 MW of capacity per Starship launch.

你把数字倒推回去,大概能得到每次 Starship 发射 5 兆瓦的算力。

There's 100 metric tons in one of those Starships.

那种 Starship 一次能装 100 公吨。

So, you can back into the math of how much will it cost per gigawatt to launch these satellites into space.

所以你可以倒推出这笔账:把这些卫星发射上太空,每吉瓦要花多少钱。

Launch this compute into space.

把这些算力发射上太空。

Um and the math that you get to before you account for things like bad GPUs, bad satellites, right, these will all be things that happen.

而你算出来的这笔账,还没算上坏 GPU、坏卫星这些——这些肯定都会发生。

But the math you get to is it's about $5 billion per gigawatt of CapEx to put these in space.

但你算出来的是:把这些放上太空,大概是每吉瓦 50 亿美元的 CapEx。

For comparison, terrestrially, talk about the switch gears, the generators, the transformers, the shell, getting the power, that today is about 25 20 to 25 billion per gigawatt.

做个对比,地面上,那些开关柜、发电机、变压器、外壳、把电搞到手,今天大概是每吉瓦 200 到 250 亿美元。

So we're talking about a 5x reduction in cost on half of your bill of materials

所以我们说的是,在你物料清单的一半上,成本降到原来的五分之一——

for the data center.

就是数据中心这部分。

Which is a huge number.

这是个巨大的数字。

Yeah, just just very simply, I mean just to say that put it cost $60 to put a on the ground today.

对,很简单地说,就是今天在地面上建一个要花 600 亿美元。

And we'll call it 35 of that is are the GPUs and the silicon that's doing the training and the inference.

我们就算其中 350 亿是做训练和推理的 GPU 和硅片。

And 25 billion is the land, the shell, the power, and the cooling.

另外 250 亿是土地、外壳、电力和冷却。

I would hypothesize that those elements are probably going to be inflationary, so that 25 billion may not go down.

我大胆假设,这些部分很可能会通胀,所以那 250 亿未必会降。

And because space, power, cooling are effectively free in space, and when I say space, I mean land.

而因为在太空里空间、电力、冷却基本是免费的,我说"空间"的时候,其实指的是土地。

You know, there's no land in space, but there is space. you're you're talking about putting a gigawatt into space for 30 billion and having lower operating costs.

太空里没有土地,但有空间。你说的是花 300 亿美元把一吉瓦放上太空,而且运营成本更低。

Now the dynamic versus 60 billion that's inflationary, and that third and that 30 billion, that five may be deflationary over time.

现在对比一下:600 亿那边是会通胀的,而那 300 亿里的那 50 亿,随着时间可能还会通缩。

But what we need to consider is you know, the reliability and the maintenance.

但我们需要考虑的是可靠性和维护。

And so as long as you know, everybody can do the math, but as long as these satellites in space aren't failing at an at an astronomical rate, the math maths.

所以只要——大家都会算这笔账——只要太空里这些卫星的故障率不是高得离谱,这账就算得平。

As you can see, and by the way, we know GPUs melt and lasers fail.

你看得出来,而且顺便说一句,我们知道 GPU 会烧、激光器会坏。

We know this happens in data centers, particularly during big training runs.

我们知道数据中心里就会发生这种事,尤其是在大规模训练跑批的时候。

And yeah, I mean GPUs melt.

对,GPU 是会烧。

Um so as long as the reliability and maintenance is not dramatically lower, the math is there once we have reusability and then rapid reusability for Starship V3.

所以只要可靠性和维护不是差得太离谱,一旦我们有了可复用性、再有了 Starship V3 的快速可复用性,这笔账就成立。

I when you when we look at this, okay, so we we went through Starlink and we said, "Okay, like it it just stands to reason we're going to have direct to cell on Starlink." Like the assumptions there are, you know, again, seem like you can get your head around.

我们回头看这个,好吧,我们先过了 Starlink,说,行,那按理说 Starlink 上会有直连手机。那些假设你还是能想明白的。

Then when it comes to building terrestrial data centers, again, not a hard one to think that based on these couple deals that Elon's going to build a much bigger Starlink's going to build or SpaceX is going to build a much bigger business there.

然后说到建地面数据中心,基于这几笔交易,不难想到 Elon 会建、Starlink 会建、或者说 SpaceX 会在那块建起一个大得多的业务。

And then you have this call option on space that would drop the price even further.

然后你还握着一个太空的看涨期权,能把价格再往下压。

The one thing we haven't talked about is their model, right?

唯一还没聊的,是他们的模型,对吧?

And I find this surprising, right?

我觉得这点挺意外的。

Six Six months ago, x.ai was competing, they were doing pretty well, but they've done something dramatic over the course of the past couple couple months, which is they bought Cursor, right?

六个月前,x.ai 还在竞争,他们做得挺好,但过去这一两个月他们干了件很猛的事,就是买下了 Cursor,对吧?

Cursor is 700 800 people was already doing incredibly well from a revenue perspective.

Cursor 有七八百人,从营收角度看已经做得极其出色。

Our own projections were that they could exit this year at up to $10 billion of revenue, so they were growing very fast one of the leading coding agents, but they also had this incredible team with the potential, right, to really build a frontier level model, but they were compute constrained.

我们自己的预测是他们今年年底营收能到 100 亿美元,增长非常快,是头部的编程智能体之一,但他们还有一支了不起的团队,有潜力真正做出前沿级别的模型,只是受算力约束。

So all of a sudden, they get bought by X.

于是突然之间,他们被 X 收购了。

X has massive compute that they can now train on.

X 有海量算力,现在他们可以拿来训练。

And when I think about the revenue in AI that like if I look at that line item in the models having it go from $10 billion to $150 billion, yes, a lot of that will be the core weave type business that they have, but the question is how much of that is going to be the core x.ai business that's really powered by the new team from Cursor.

我在想 AI 里的营收,如果我看模型里那一行,从 100 亿涨到 1500 亿美元,是的,里面很大一块会是他们那种 CoreWeave 类型的业务,但问题是有多少会是真正靠 Cursor 那支新团队驱动的 x.ai 核心业务。

So any thoughts on that, Kevin?

所以你怎么看,Kevin?

Right now, so Composer 2.5 was Pareto dominant 12 days ago.

现在,Composer 2.5 在 12 天前是 Pareto 占优的。

It was trained on the Kimmy K2.5 base model.

它是在 Kimi K2 基座模型上训练的。

Now, what's happening is the Grok 4.3 1.5 trillion parameter model is training.

现在正在训练的是 Grok 4.3、1.5 万亿参数的模型。

One would hypothesize based on scaling laws that that will might be a better base model.

按照 scaling laws 推测,那可能会是个更好的基座模型。

And then the cursor data is being injected into the pre-training process, not just reinforcement learning.

然后 Cursor 的数据被注入到预训练过程里,而不只是强化学习。

And we'll see, and I think that is going to be a very important data point when that comes out.

我们拭目以待,我觉得等它出来会是个非常重要的数据点。

And I just think everyone should keep in mind that once you are at multiple places on that Pareto curve, if you have compute, you can scale really rapidly.

我觉得每个人都该记住,一旦你在那条 Pareto 曲线上占了好几个点位,只要你有算力,就能非常快地扩张。

You know, that that to me is if I had to say what the one piece that's being lost in the story, right?

对我来说,如果要说这个故事里被忽略掉的那一块是什么,就是这个。

Like it's easy for everybody to get excited about the deals with Anthropic because you can put your hands around that.

大家很容易为跟 Anthropic 的交易兴奋,因为那个你抓得住。

You know how much revenue it is.

你知道那是多少营收。

I see debate about, you know, the 90-day termination and how long they last and what multiple do you put on those revenues.

我看到有人在争论那个 90 天终止条款、能撑多久、这些营收该给多少倍估值。

But I think the thing that's getting lost is I think they've dramatically advanced their capability when it comes to building a frontier model.

但我觉得被忽略的是,在做前沿模型这件事上,他们的能力已经大幅跃进了。

People outside Silicon Valley may not know, you know, Michael and the team at Cursor as well.

硅谷之外的人可能不太了解 Cursor 的 Michael 和他的团队。

This is an extraordinary team that he just downloaded, right, into SpaceX.

这是一支了不起的团队,他刚整个搬进了 SpaceX。

SpaceX was already building good models.

SpaceX 本来就在做不错的模型。

And what they have is they have this way to monetize compute that gives you this call option that you can pull all that compute in-house, right, to train a model and then to run the model.

他们手里的东西,是有这么一套变现算力的方式,给了你这么一个看涨期权:你可以把所有那些算力全都收归内部,去训练模型、再运行模型。

I suspect if there's an upside surprise, if we went around the table, I'd say this is the place that's getting the least amount of attention and could have the biggest upside surprise.

我猜如果有超预期的惊喜,如果我们绕桌子一圈问下来,我会说这就是那个受关注最少、却可能带来最大超预期惊喜的地方。

Any any thoughts, Clark, on what you think is being overlooked or areas that you think are misunderstood about the business today?

Clark,你有什么想法,你觉得现在这个业务里有哪些被忽视、或者你觉得被误解的地方?

I I would say I would say what the last few weeks have proven is that Elon, um, their team can stand up all this compute.

我会说,过去这几周证明的一点是,Elon 和他们的团队能把这么多算力搭起来。

Actually, if you just, you know, went back 1 and 1/2 years, you know, they were behind in the race to stand up compute.

其实你要是往回倒一年半,他们在搭算力这场竞赛里还是落后的。

They were you know they they didn't have that many H100s.

他们当时没有那么多 H100。

They brought in Colossus.

他们搞来了 Colossus。

Then they brought in Colossus 2 at a scale much larger than anyone else.

然后又搞来了 Colossus 2,规模比其他任何人都大得多。

And now you know as we gear for Vera Rubin you know from you know a lot of my conversations it looks like they've you know secured maybe up to 20% of Vera Rubin capacity especially in the early days of you know when you know these these chips are very scarce that that they're going to have a a a lead on all of this because you know people think that they can stand up this compute better.

现在我们为 Vera Rubin 做准备,从我很多次对话来看,他们似乎锁定了 Vera Rubin 差不多两成的产能,尤其在这些芯片非常稀缺的早期,他们会在这一切上领先,因为大家觉得他们更擅长把这些算力搭起来。

So I think they'll all you know what what the last few weeks have actually shown is that Elon you know Elon will take you know take a shot at hitting the frontier but if it you know if for whatever reason um they they have over procured some capacity this is a very scarce asset that they've shown that they can monetize at actually you know best in class margins and payback periods.

所以我觉得,过去这几周真正显示出来的是,Elon 会去冲一冲前沿,但如果不管出于什么原因他们过量采购了一些产能,这是个非常稀缺的资产,他们已经证明自己能以业内一流的毛利率和回本周期把它变现。

The irony is like you know you and I've been doing this long enough to know I mean that's why Bezos built AWS.

讽刺的是,你我干这行够久了都懂,我是说,这正是 Bezos 当年建 AWS 的原因。

Right?

对吧?

He had to build capacity for Black Friday.

他得为黑色星期五建产能。

Right?

对吧?

But then the rest of the year he sat on all this capacity they had to build and he figured out a really incredible way to monetize this.

但一年里其余时间,他守着这堆不得不建的产能,然后想出了一套非常惊人的变现办法。

And by the way investors at the time 2009 2010 when he was building out the capability around AWS hated it.

顺便说一句,当年 2009、2010 年他在搭 AWS 那套能力的时候,投资人都讨厌这事。

Of course.

当然。

Because he was consuming all that free cash flow.

因为他把那些自由现金流全烧掉了。

My meanwhile he was digging the biggest gold mine in the history of the world.

与此同时,他正在挖世界史上最大的一座金矿。

One of the biggest.

之一。

One of the biggest.

最大之一。

Among them among them at the time was probably the biggest.

在当时,它们当中大概是最大的。

Yeah Google search might want to have a we'll have a we'll have a discussion.

是啊,Google 搜索可能要来跟我们聊聊了。

By the way I do think it is important.

顺便说一句,我确实觉得这点很重要。

Grok 4.3 I think the cursor if they acquire it that may end up being very important.

Grok 4.3,我觉得 Cursor,如果他们收下来,那最后可能会非常重要。

But Grok 4.3 was on the Pareto frontier and has of 10 or 12 days ago and this these things move fast.

但 Grok 4.3 在 Pareto 前沿上,这是十天或十二天前的情况,而这些东西变化很快。

But most intelligent 500 billion parameter model in the world.

但它是全世界最聪明的 5000 亿参数模型。

And they were on the frontier and there are four companies on the frontier. xAI, SpaceX AI, Google one with Gemini 3.1 Pro, and then the rest of it was dominated by Anthropic and OpenAI.

他们在前沿上,而前沿上有四家公司:xAI、SpaceX AI、Google 靠 Gemini 3.1 Pro 一家,剩下的则由 Anthropic 和 OpenAI 主导。

But they were on the Pareto frontier and now we'll see what they do with Cursor.

但他们在 Pareto 前沿上,现在就看他们拿 Cursor 做出什么了。

Yeah.

对。

Um I want to come back to that in a second.

这个我想过会儿再回来聊。

way, man, I want to ask you some questions.

等下,老兄,我想问你几个问题。

go go go.

来来来。

What do you think?

你怎么看?

So you think the biggest source of potential upside is the model?

所以你觉得潜在上行空间最大的来源是模型?

What do you think?

你怎么看?

I think that's the I think that's the thing that's least talked about.

我觉得这恰恰是最少被人讨论的一块。

Least talked about.

最少被讨论。

Right?

对吧?

And so, listen.

那你听我说。

When I look at the bull bear case on the IPO, right?

我看这次 IPO 的多空双方观点,对吧?

The bears are looking at last year's revenue.

空头盯的是去年的营收。

Say it was $18 billion and they're looking at the forecast from the banks of $160 billion, you know, 3 years from now and they're saying, "Listen, not many companies in the history of the world have basically 8x their revenue over 3 to 4 years." Right?

假设去年是 180 亿美元,他们再看投行给出的三年后预测——1600 亿美元,然后说:“听着,世界历史上没几家公司能在三四年里把营收翻上 8 倍。”对吧?

So that's where, you know, I think and people get nervous about the valuation.

所以这就是让大家对估值发怵的地方。

When I look at this, again, when you break it down as an analyst first principles, part by part, which is what I tried to do here, right?

而我看这件事,还是那句话,当你像分析师那样按第一性原理一块一块拆开——这正是我在这儿想做的,对吧?

When you look at Starlink, it looks totally doable.

你看 Starlink,完全做得到。

When I look at what they're building in AI compute terrestrially, looks totally doable over the course of next 3 years.

我看他们在地面上搭建的 AI 算力,未来三年也完全做得到。

When I look at the model itself after the acquisition of Cursor, you know, combining those things around the compute they have, that looks to me like it could be an upside surprise.

我再看模型本身,收购了 Cursor 之后,把这些东西跟他们手里的算力结合起来,在我看来这块很可能会给人一个上行的惊喜。

So I would say that I think that uh you know, in the IPO, but I think when you look back 3 years from now, there's a decent chance that everybody's like, "Oh my god, that was super obvious." Right?

所以我会说,IPO 当下,但等你三年后回头看,很有可能大家都会说:“天哪,这不是明摆着的吗?”对吧?

Even though today all of these things have risk associated and back to where we started.

尽管今天这几件事都各带风险,又回到我们最开始那句话。

I'm not, you know, none of us are here to pump the IPO at 1.77 trillion.

我不是,我们谁都不是来给这个 1.77 万亿美元的 IPO 抬轿子的。

It's really to just break it down as we do inside our shop and to say, "What is that distribution of future probabilities?

真正要做的,就是像我们在自家机构里那样把它拆开,然后问:“未来概率分布长什么样?

What's the probability that it's higher from here?

从这儿往上走的概率有多大?

What's the prob" And I think we're all pretty AI pilled.

概率是多少”——而我觉得我们几个都挺信 AI 那套的。

And if you're AI pilled, that means we got to build a lot more compute than the world thinks and that these models are going to be a lot more valuable than people think.

而如果你信了 AI 那套,那意味着我们得建的算力要远超世界的预期,而这些模型的价值也会远超大家的想象。

You combine that with their core business.

你再把这一点跟他们的核心业务结合起来。

I don't know another entrepreneur or another business that's a better bet on the future, forget it, right, in order to have a

我想不出还有哪个创业者、哪门生意,是押注未来更好的一注,别提了,对吧,为了拥有一个

From your lips to God's ears.

但愿如你所说。

I mean, listen, I I again, I think that I I I think that you're going to have to wait, but you know, we had this chart last week, right, that came out.

我是说,听着,我还是那句话,我觉得你得等一等,不过我们上周有张图,对吧,刚出来那张。

Everybody was sending around Twitter, conveniently timed, and you know, it's like shows the average max drawdown post IPO for like 20 companies from Facebook, Twitter, Alibaba, Shopify is, you know, over 50%.

大家都在 Twitter 上转发,时间点还挺凑巧,那张图显示 Facebook、Twitter、阿里巴巴、Shopify 这 20 来家公司 IPO 之后平均最大回撤超过 50%。

And so maybe that again will will will end this section here.

所以也许还是这样,这一段我们就在这儿收尾吧。

You know, Gavin, you and I've been doing this a long time.

你知道,Gavin,你和我干这行都很久了。

We know it's going to be bouncy around the IPO.

我们清楚 IPO 前后会很颠簸。

Um, you know, how do you as a manager try to try to manage that?

那你作为管理人,会怎么去应对这种情况?

Um, do you try to trade around the IPO?

你会围绕 IPO 做波段交易吗?

Do you set it kind of and forget it?

还是买了就放着不管?

I would say from an Altimeter perspective, what we tend to do is we take a base position that we set and forget, right?

我会说,从 Altimeter 的角度,我们通常的做法是建一个底仓,买了就放着不动,对吧?

And then we may size up or size down depending upon how the market reacts in, you know, in a particular moment.

然后我们可能会根据市场在某个具体时点的反应来加仓或减仓。

Um, but any thoughts on on this chart or you know, how how people you guys are thinking about it in particular.

那对这张图你有什么想法,或者你们几个是怎么看的。

You obviously own a lot going into it.

你们显然在 IPO 前就持有不少。

First agree with absolutely everything you said and I actually think about it the same way, set it and forget it.

首先我完全同意你说的每一句,而且我其实也是这么想的,买了就放着不用管。

You've talked about you have ballast, you move around and you move the ballast to one side of the ship when you want to the ship to lean into the wind to go faster and you move it to the other side when you don't want the ship to tip over.

你之前讲过,你有压舱物,你会来回挪,当你想让船顶着风开得更快时就把压舱物挪到一侧,当你不想让船翻时就挪到另一侧。

I think that's a great analogy.

我觉得这个类比很棒。

Think about all important companies in the portfolio the same way.

我对投资组合里所有重要的公司都是这么看的。

So 100% agree.

所以百分之百同意。

I mean, this this chart is a bummer.

我是说,这张图真让人扫兴。

What I would say is, you know, this data on IPOs, but what I would just say is this is a really unprecedented situation.

我想说的是,这些 IPO 数据,但我只想说,这是一个非常史无前例的局面。

Yes.

是的。

We've never had an IPO this big.

我们从没见过这么大规模的 IPO。

We've never had an IPO that's going to go into an index this quickly.

我们从没见过一家公司这么快就要被纳入指数。

We simply do not know how much selling there will be from investors.

我们根本不知道投资者会有多大的抛售量。

I would hazard a guess.

我斗胆猜一下。

I mean, I'm I don't know.

我是说,我也不知道。

But Elon, I don't think he needs liquidity and I think he owns What does he own, Foxy?

但 Elon,我觉得他不需要流动性,而且我觉得他持有——他持有多少来着,Foxy?

It's 50%

是 50%

50% of the company.

公司的 50%。

way, he's locked up for 365 days or 366 days.

而且他被锁定了 365 天或者 366 天。

So, we know he's not selling, right?

所以我们知道他不会卖,对吧?

So, I just think it's an unprecedented situation and the right answer

所以我就觉得这是个史无前例的局面,而正确的答案

is I don't know what's going to happen in the short term.

就是,短期内会发生什么我也不知道。

And the right answer that I would just, you know, encourage every investor making their own decision is to just think exactly [clears throat] the way you articulated it.

而我想鼓励每一位自己做决定的投资者去采用的正确答案,就是完全按你刚才讲的那个思路来想。

We have these different levers.

我们有这些不同的杠杆。

We have these different variables.

我们有这些不同的变量。

Think about each one of them from first principles.

从第一性原理去思考每一个。

Make your own decision.

自己做决定。

Do your own due diligence.

自己做尽职调查。

Be thoughtful.

想清楚。

But, there are a lot of variables here and that it is a little funny to me that uh you know, it was 100 times trailing TTM revenue.

但这里有很多变量,而且有件事我觉得还挺好笑的,就是这价格曾经是过去十二个月 TTM 营收的 100 倍。

Well, after the deals they signed, I think it's at 39 times.

而在他们签了那些合同之后,我觉得现在是 39 倍。

That can change fast.

这个可以变得很快。

So, they added $29 billion in a month.

所以他们一个月里增加了 290 亿美元。

Yes.

对。

Now, it's

现在,它

[laughter]

[笑]

By the way, have you ever seen that happen?

顺便问一句,你见过这种事发生吗?

Never.

从没。

Never.

从没有过。

And you know, it just goes to show first um Elon is not only a great engineer.

你知道,这恰恰说明,第一,Elon 不只是个了不起的工程师。

He and Gwen and the team are great at business.

他、Gwynne 还有整个团队,做生意也是一流的。

And Brad,

而且 Brad,

They they they understand what needs to be done to raise the capital to get to the next phase.

他们非常清楚,要进入下一阶段该做什么、该怎么把资本融到位。

They have a long-term mission in the business.

他们在业务上有一个长期使命。

And so, to me, again, what we saw in the course of the last few weeks with cursor, what we saw with these deals that they cut, I don't know that any of the mag seven could have moved that quickly to adjust the business that they did.

所以在我看来,再说一次,过去几周我们在 Cursor 身上看到的,还有他们谈成的这些交易,我不觉得 Mag 7 里有哪家能这么快地把业务调整到位。

It's exceptionally entrepreneurial at scale, which we very rarely see in businesses.

这是在极大规模上还保持着高度创业精神,这在企业里非常罕见。

Two other things I would just say

我还想再说两点

you a hug, Brad?

抱一个吗,Brad?

Two Two other things I I I I would just say.

我还想再补充两点。

Number one is people talk a lot about the total amount of capital being raised.

第一点,大家老是在谈融资的总规模。

If you add up the capital here, right, for Anthropic what they may raise, what OpenAI may raise, what, you know, SpaceX may raise, let's call it $250 billion.

你把这里的资本加总起来,Anthropic 可能融的、OpenAI 可能融的、SpaceX 可能融的,就算它 2500 亿美元。

That's 1% of the Mag 7.

那也就是 Mag 7 的 1%。

Okay, it's 1% of the Mag 7.

好吧,是 Mag 7 的 1%。

I Yeah.

我,对。

And And we will as well.

而且我们也会跟进。

You know, like that to me is like a bet on the future that we all believe in.

你知道,这在我看来就像是押注于一个我们都相信的未来。

And so, if I said, "Where are we out of consensus?

所以,如果要我说,我们在哪儿跟市场共识不一样?

What is our variant perception?" We actually think it's going to be bigger, faster, and we've thought that for a couple years.

我们的差异化认知是什么?我们其实认为它会更大、更快,而且这个判断我们已经持有好几年了。

Um so, first, it's only 1% of the Mag 7 market cap.

首先,它只占 Mag 7 市值的 1%。

And then you referenced it, the amount of selling.

然后你刚提到的,抛售的规模。

Um I've got a chart we'll post here.

我这儿有张图,我们会贴在这里。

This is, you know, the the dribble share release for SpaceX shareholders.

这是 SpaceX 股东那种一点一点释放的份额。

You know, so there's not a lot that can be released um up until after the first earnings.

你知道,在第一次财报出来之前,能释放的量并不多。

This We saw this in the Cerebras IPO.

这个我们在 Cerebras 的 IPO 里就见过。

Um there's a version of it here in this IPO.

这次 IPO 里也有一个类似的版本。

And so, again, I think the banks have been thoughtful here, knowing that this is a very large IPO.

所以再说一次,我觉得投行在这件事上考虑得很周到,他们知道这是一次非常大的 IPO。

And I'm not saying that won't trade down.

我不是说它不会跌。

Like there's possibility, you know, these things trade down.

有这个可能,你知道,这类股票会往下走。

But again, for me, telescope out, is there any company better positioned as a bet on the future?

但对我来说,再把镜头拉远一点,有没有哪家公司作为对未来的押注,处在比它更好的位置?

I think what they've shown over the course of last 5 weeks, they're they're they're probably number one.

我觉得过去五周他们所展示出来的,大概率是第一名。

But let's move on.

但我们往下说吧。

No, no, can I just say one thing about the employees?

不不,我能不能就员工的事说一句?

I think another thing that's unprecedented here is the employees

我觉得这里另一个前所未有的地方,是员工

and to a large degree the investors here have had liquidity every 6 months.

还有很大程度上这些投资人,每六个月就能有一次流动性。

Exactly.

一点没错。

the last 10 years.

过去十年都是这样。

So, if you're a SpaceX employee or former employee, and you wanted to sell you've had whatever that is, close to 20 chances.

所以,如果你是 SpaceX 的员工或前员工,你想卖,你差不多有过将近 20 次机会。

And it is a matter of historical record that large investors have been able to sell.

而且大投资人一直能卖出,这是有历史记录可查的事实。

So I would think a lot of the people

所以我会觉得,很多人

they've

他们

chosen to own it.

是主动选择继续持有它的。

Now, there's a new valuation and we'll see what they do, but just this is utterly unprecedented and we'll see.

现在有了新的估值,我们再看他们会怎么做,但这实在是前所未有,我们拭目以待。

Yeah, I know.

是,我知道。

It's It's It's a great point.

这是个很好的点。

We have in fact called these companies quasi-public.

我们其实就把这些公司叫做准上市公司。

Um you and I both know that SpaceX and I'd put Anthropic in in in this category as well, Databricks in this category.

你我都清楚,SpaceX,还有 Anthropic 我也会归到这一类,Databricks 也在这一类。

These things in many ways have been more liquid over the course of the past 3 years than some public biotech companies we know.

这些东西在很多方面,过去三年里比我们知道的一些上市生物科技公司还要有流动性。

Right?

对吧?

And so there's a continuum of liquidity here.

所以这里存在一条流动性的连续谱。

We We treat it as a binary, private versus public, but it's really about this continuum.

我们习惯把它当成非黑即白,私有对上市,但真正的关键是这条连续谱。

You know, let's keep going on models.

我们接着聊模型吧。

You know, um Anthropic launched Fable 5, which you referenced um yesterday, which is basically Mythos um with some classifiers and safeguards um around cyber and biology, chemistry, um and distillation.

Anthropic 发布了 Fable 5,就是你昨天提到的那个,它基本上就是 Mythos,外加一些围绕网络安全、生物、化学以及蒸馏的分类器和防护措施。

When those things get triggered, it fails back [snorts] to Opus 4.8.

当这些东西被触发时,它会回退到 Opus 4.8。

Um you know, there was a Copart tweet about this yesterday.

你知道,昨天有一条 Copart 的推文谈到这个。

He said, you know, it sold on all the benchmarks, but what really makes it special is long-running tasks.

他说,你知道,它在所有基准测试上都拿下了,但真正让它特别的是长任务。

Okay?

好吧?

You retweeted our good friend, you know, Noam Brown.

你转发了我们的好朋友 Noam Brown。

Um you know, ChatGPT 5.5 also exhibited these capabilities.

ChatGPT 5.5 也表现出了这些能力。

Um you know, it it led Noam, right, to suggest that it's not very relevant to do these snapshot benchmarks anymore.

这让 Noam 提出,再去做这种快照式的基准测试已经不太有意义了。

Yeah, like the x-axis has to be time or tokens or compute because we can solve most problems now if we just let these frontier models for a very long uh point in time.

对,横轴得换成时间、token 或算力,因为只要我们让这些前沿模型跑足够长的时间,现在大部分问题我们都能解出来。

So, Gavin, what is this new class of model, right, Fable Fable 5, ChatGPT 5.5?

那么,Gavin,这个新一类的模型,Fable 5、ChatGPT 5.5,

What does it mean for the race in superintelligence?

对超级智能这场竞赛意味着什么?

Who's up?

谁领先?

Who's down?

谁掉队?

Who's still on the frontier?

谁还在前沿上?

Um give us your thoughts.

说说你的想法。

I mean, it's hard to say that Anthropic's not up.

我是说,很难讲 Anthropic 没起来。

Like after the revenue numbers they've put up, after the Fable 5 release, and Mythos is evidently even better.

看看他们放出来的营收数字、Fable 5 的发布,而 Mythos 显然还要更强。

But I just think that Gnome Brown post from yesterday, polynomial, is so profound.

但我就是觉得,Noam Brown 昨天那条帖子,polynomial 那条,太深刻了。

And just the idea that we do not know how smart these models are.

就是那个观点——我们并不知道这些模型到底有多聪明。

And we made

而我们做出来的——

Say more about that.

展开讲讲。

Why don't we know how smart they are?

为什么我们不知道它们有多聪明?

Because nobody has run Mythos for a year continuously.

因为没人把 Mythos 连续跑过一整年。

And we may never know how smart each generation of models actually is or was, but because we don't have time to appropriately evaluate their intelligence before the next model comes out.

我们可能永远不会知道每一代模型实际有多聪明、曾经有多聪明,因为下一个模型出来之前,我们根本没时间好好评估它的智能水平。

I mean, this is a profound statement.

我是说,这是个非常深刻的论断。

And just just imagine, okay?

你就想象一下,好吗?

So, I always say like when you think about FSD, just imagine a human being who never gets distracted, never gets tired, never talks on the phone in the car, never drinks and drives, never yells at their kids, never has to go to the backseat to give their baby a bottle.

我老是说,想想 FSD,想象一个人从不分心、从不疲劳、开车时从不打电话、从不酒驾、从不冲孩子吼、也从不为了给宝宝喂奶而钻到后座去。

And like of course you would think that over time that is superior to humans who are distracted.

那你当然会觉得,时间一长,这肯定强过会分心的人类。

I don't know how long How long can you think deeply about one topic, Brad?

我不知道能持续多久——Brad,你能对一个话题深度思考多久?

What do you Give me an hour.

你能——给我一小时。

Give me an [laughter] hour.

给我一[笑]小时。

Give me an hour.

给我一小时。

A BIT.

就一点点。

THAT MAKES me feel terrible cuz I think I can think deeply about one topic continuously before having a stray thought enter my mind for like maybe 5 minutes.

这让我特别难受,因为我觉得自己对一个话题连续深度思考、脑子里不冒出杂念,大概也就能撑五分钟。

Then I can come back to that.

然后我才能再回到那个话题上。

Imagine if Albert Einstein had been able instead of, you know, and maybe that maybe I maybe he could think for 3 hours at a time.

想象一下,如果 Albert Einstein 能够——也许他一次能思考三个小时。

Clearly an exceptional intellect.

显然是超凡的智力。

But imagine Albert Einstein had just thought about fundamental physics 24 hours a day.

但想象一下,Albert Einstein 一天二十四小时都在思考基础物理。

He doesn't have to eat, he doesn't have to sleep, he doesn't have to relax, he doesn't drink,

他不用吃饭、不用睡觉、不用放松、不喝酒,

never gets old,

永远不会变老,

never has diminished intelligence,

智力永远不会衰退,

and he thought for 1 year.

而且他连续思考了一年。

I mean, we might already, you know,

我是说,我们说不定已经

have solved a lot of these intractable problems.

解决了很多这些棘手的难题了。

So, I just think that's an extraordinary thought.

所以我就觉得这是个了不起的想法。

And just my takeaway was however bullish I was on compute before then, I'm just a lot more bullish.

我的结论是,不管我之前对算力有多看好,现在我只会更看好得多。

Right.

对。

Right.

对。

Right.

对。

So, so, so that is a, you know, we saw when that was probably what really unlocked Opus 4.6.

所以这就是——你知道,我们看到,那大概就是真正解锁 Opus 4.6 的东西。

It was the first really long-running model that could maintain that context, maintain that memory, um solve some of these longer-running problems, right?

它是第一个真正能长时运行的模型,能维持住上下文、维持住记忆,解决一些更长周期的问题,对吧?

For us, the signal was in January.

对我们来说,信号出现在一月。

We knew we felt like that was a big moment, but then when you started to see the revenue go up, we knew that lots of people were voting independently, that that was a profound moment that they became much, much more useful.

我们当时就感觉那是个大时刻,但等你开始看到营收往上走,我们就知道很多人在各自独立地投票,证明那确实是个深刻的时刻——它们变得有用得多、多得多。

So, but one of the things that the consensus going into this year, right?

那么,今年年初大家的共识之一,对吧?

So, the big question going into this year was was the AI revenue going to show up?

今年年初最大的问题是:AI 营收会不会兑现?

Were we going to get to these thresholds of intelligence that caused enterprises and consumers to use them more?

我们会不会达到那些智能门槛,让企业和消费者更多地用它们?

And I think the consensus at the time, at least on this podcast, um the the the debate with with my with with with Bill was the open-source models, cheap tokens, were catching up on the frontier, that perhaps these models were beginning to asymptote, um that people wouldn't really pay for premium tokens, and it seems to me that the evidence on the field, 6 months into the year, is just the opposite, right?

我觉得当时的共识,至少在这个播客上,我和 Bill 的那场争论是:开源模型、便宜的 token 正在前沿上追赶,也许这些模型开始触顶趋平,也许人们不会真的为高端 token 付费——可在我看来,今年过半、场上的证据恰恰相反,对吧?

That frontier tokens are capturing the vast majority of all the revenues, and that in fact, if you believe in the long-running capabilities and more compute allows you to do that, they may actually be extending their lead, right?

前沿 token 拿走了绝大部分营收,而且事实上,如果你相信长时运行的能力、相信更多算力能让你做到这一点,它们说不定还在拉大领先优势,对吧?

On some of these models that were built on distillation.

就是那些建立在蒸馏之上的模型。

So, I just open it up to anyone around the table, what are your thoughts on whether or not, you know, have we challenged this thesis that cheap open-source tokens are going to always, you know, close the gap on these frontier models, or are they extending their leads?

所以我把话题抛给在座各位,你们怎么看——我们是不是已经挑战了这个论断:便宜的开源 token 总会追平这些前沿模型,还是说前沿反而在拉大领先?

I I think this debate, like this same debate has existed since the beginning of since we started training these models to begin with, which was hey, we're always kind of three, six months behind the frontier.

我觉得这场争论,从我们一开始训练这些模型的时候就存在了,就是那句"我们总是落后前沿三到六个月"。

But empirically, like you can just see all of the revenue has actually just accrued at the frontier.

但从实证看,你能清楚看到,所有营收其实都汇聚到了前沿上。

And that I think that's because every time we release the frontier, a whole new like slew of use cases

我觉得那是因为,每次我们发布前沿模型,就会冒出一大批全新的用例,

that that previously we could have never tackled before, like coding.

那些以前我们根本没法碰的东西,比如写代码。

Um but also just, you know, you know, we've we've just been locked at our desk for the last last day just, you know, hammering Claude because, you know, it's just fascinating the things that now we can do with fable five that we could just couldn't do with opus 48 just a day before.

而且说真的,我们过去这一天就锁在桌前疯狂敲 Claude,因为太让人着迷了——用 Fable 5 现在能做到的事,就在一天前用 Opus 48 还根本做不到。

So what are some of those things, man?

那都有哪些事啊,老兄?

I'm curious.

我很好奇。

So So I think it's really really good at multi-agent orchestration.

我觉得它在多智能体编排上真的非常强。

So they they Anthropic released a um a blog post about like different uh agent um six different agent like orchestration patterns that, you know, they've they've talked about.

Anthropic 发了一篇博客,讲了六种不同的智能体编排模式,他们聊过的那种。

But really like once you start being able to manage all these agents, the harness and the model itself is being arled with one another, they're actually being, you know, fused closer and closer together, but the model can understand the, you know, the extent of your work.

但真正一旦你开始能管理这一堆智能体,harness 和模型本身就在彼此磨合、越来越紧密地融合到一起,而模型能理解你工作的全貌。

So, you know, one of the things, for instance, is um I just threw in like seven of our models and just said, "Okay, like I want to create a master view of like my beliefs given all of these assumptions of all these companies, TSMC capacity, like and then and then produce me a report on all this stuff." And you know, the the model is able to reason through all of our assumptions.

举个例子,我随手扔进去七个我们的模型,然后说:"好,我想基于所有这些公司的假设、TSMC 产能之类,建一个我所有观点的总览视图,再给我产出一份关于这一切的报告。"而模型能把我们所有假设都推理一遍。

Like actually, if you believe this

就是那种,如果你相信这个——

Right.

对。

What are the contradictions exactly?

到底有哪些矛盾之处?

Yeah, it's it was fascinating.

是啊,太让人着迷了。

And and and you know, before we'd never do that, but but now, you know, I think we're just step one into multi-agent orchestration.

你知道,以前我们绝不会那么干,但现在,我觉得我们才刚迈进多智能体编排的第一步。

We're going to do this even further and that's one example.

我们还会把这个推得更远,这只是一个例子。

I've also dumped all my all my notes into it and it's reason across all my notes from the last 3 years and said, you know, here are some of your ideas that were consistent.

我还把我所有笔记全倒了进去,它把我过去三年的笔记通读了一遍,然后说:这是你一以贯之的一些想法。

Here are like, you know, the sources that were actually the highest signal to what actually played out, you know, and then it is actually just super fascinating what you could do and we've just blown through our blown through our limits.

这是——你知道,那些对最终真正发生的事信号最强的来源,然后你能做到的事真的超级让人着迷,我们已经把额度冲爆了、冲爆了。

mean it's it's it's unlocking all this.

我是说,它把这一切都解锁了。

I mean like they gave examples yesterday and the release Anthropic did, you know, 50 million line Ruby code base at Stripe that was, you know, refactored in a day versus many weeks with many people.

我是说,他们昨天发布里就举了例子,Anthropic 在 Stripe 那个 5000 万行的 Ruby 代码库,一天就重构完了,而过去要很多人干很多周。

You think about where this is impacting biology and life sciences just across the spectrum. and to me it really gets back to this fundamental point.

你再想想这对生物学、生命科学的影响,横跨各个领域。对我来说,这真的又回到了那个根本点上。

Number one, if you believe this to be true about long-running agents, then we're going to produce and consume more tokens in the future as far as the eye can see.

第一,如果你相信长时运行的智能体这件事是真的,那我们未来会生产和消耗越来越多的 token,一眼望不到头。

So the world this gets me back to, you know, terrafab and space orbital and all this because we we we may in fact unlock real thresholds of intelligence, but we're going to have to let these horses run for a long time in order to get there.

所以这又把我带回到 terrafab、太空轨道这些上面,因为我们说不定真的会解锁真正的智能门槛,但要走到那一步,我们得让这些"马"跑上很长一段时间。

Yeah, I would just say two things I two things can be true.

是啊,我就说两点——两件事可以同时成立。

The majority of economic value may continue to accrue to the frontier and man has it ever accrued to the frontier thus far and for sure the first 6 months of this year, but the majority of tokens consumed in the world may be open source.

大部分经济价值可能会继续归于前沿模型——目前为止确实一直归于前沿,今年头六个月尤其如此——但世界上消耗的大部分 token 可能是开源模型。

And they are

而且它们确实——

today.

今天就是。

Yes.

是的。

I and I think that this current state is likely to persist.

我觉得眼下这种状态很可能会持续下去。

Harvey had a great blog post that they put out on X and they used and it's just amazing how everything gets out out of date like in 5 days, you know.

Harvey 在 X 上发过一篇很棒的博客,他们用了——太惊人了,一切东西差不多五天就过时了。

But they used their own proprietary legal data to do reinforcement learning and supervised fine-tuning with Fireworks on an open source model and then And used a router and a router being something that picks which model you send which query to, and which model you use to check which model.

但他们用自己专有的法律数据,配合 Fireworks 在一个开源模型上做强化学习和监督微调,然后用了一个 router——router 就是决定把哪条查询发给哪个模型、以及用哪个模型去检查哪个模型的东西。

And they got better outcomes than Opus 4 either 4.7 or 4.8 at a lower cost.

结果他们拿到了比 Opus 4(不管是 4.7 还是 4.8)更好的效果,成本还更低。

And I think that is the future.

我觉得这就是未来。

And the reality is they were still consuming a lot of Opus, but a majority of the tokens they were processing probably were in their own open-source models.

现实是他们仍然消耗了大量 Opus,但他们处理的大部分 token 很可能跑在自己的开源模型上。

We heard the same thing.

我们听到的是同样的说法。

We did a We did We did an enterprise survey that we'll post of 300 companies how which ones were optimizing, so these are folks who are kind of looking at model routing and saying we're going to send certain tokens over here, which ones are thinking about optimizing, which ones aren't optimizing yet, and then what is their expected use of frontier model tokens, right?

我们做过一份企业调研,回头会发出来,覆盖 300 家公司——哪些在做优化,也就是这些人在研究模型路由,说我们要把某些 token 发到这边,哪些在考虑优化,哪些还没优化,然后他们预期会用多少前沿模型的 token?

And they're all expecting to consume a lot more even though they're already in the process of optimizing.

结果他们全都预期会消耗多得多,哪怕已经在优化的过程中了。

Think of it in the in in the context of JP Morgan.

拿 JP Morgan 来想这件事。

If they're doing some back of the house stuff, right, on customer service or whatever, they may very well use an open-source model.

如果他们在做一些后台的活儿,比如客服之类的,那他们很可能会用开源模型。

Now, I think they're loath to use Chinese open-source models, so they're waiting on kind of US open-source models to, you know, be able to really deliver the bang that they need, but my hunch is for these enterprises, a lot of that back of the house stuff will get rooted there.

现在我觉得他们不太愿意用中国的开源模型,所以在等美国的开源模型能真正拿出他们需要的效果,但我的直觉是,对这些企业来说,很多后台的活儿会被路由到那边。

That will probably be a majority of the tokens, but I think the really high-value stuff, you know, coding as an example, they don't want to write second-tier code.

那大概会占大部分 token,但我觉得真正高价值的东西,比如写代码,他们不想写二流的代码。

I think the vast majority of that will continue to be on

我觉得那部分绝大多数会继续留在——

Um

呃——

You don't need Albert Einstein to book you a trip.

你用不着 Albert Einstein 帮你订个行程。

You don't need Albert Einstein to do KYC.

你用不着 Albert Einstein 去做 KYC。

But but but this is the debate we had at literally at this table two years ago.

但这正是我们两年前就在这张桌子上争论过的话题。

However, if you just look at the revenue curves, right?

可是,如果你光看营收曲线呢?

What bill What what folks concluded when they said that, they said, "Therefore, the frontier models will not accrue most of the revenue." And what we're seeing right now, it's 90% of the

当时那些人得出的结论是,他们说:"所以前沿模型不会拿走大部分营收。"而我们现在看到的是,90% 的——

That has been decisively wrong.

那个结论已经被彻底证伪了。

Probably more than 90%, and it may continue to be decisively wrong.

可能不止 90%,而且它可能会继续被彻底证伪。

Frontier might be 90% of the economic value.

前沿模型可能占 90% 的经济价值。

Open-source

开源模型——

might be 80% of tokens.

可能占 80% 的 token。

Something that I think is very important on open source is that you know, I think there's this belief that it's bearish for AI.

关于开源,我觉得有一点非常重要,就是大家有种看法,认为它对 AI 是利空。

It's actually it may be very bearish for the frontier models.

其实它可能对前沿模型是非常大的利空。

There's that bear case you talked about.

就是你刚说的那个看空理由。

It's actually really bullish for compute and hardware because if the frontier models are capturing less of the margin, then you're going to spend more on compute.

但它对算力和硬件其实是非常大的利好,因为如果前沿模型拿走的利润变少了,你就会在算力上花更多钱。

So, the better open source does, the better it is for compute providers.

所以开源做得越好,对算力供应商就越好。

And I yeah, I I will say it there is a very I would say between um spending time in the heart of like the West, Silicon Valley, and also spending time in Asia, there is like a very big um like a deep-seated belief in one versus the other, which is like if you spend a lot of time here, it's like all closed source, cloud, every all traffic is going to go, you know, by way of this direction.

我得说,这一点很——我会说,我既在西方的核心也就是硅谷待过,也在亚洲待过,两边对这一个还是另一个有种非常大的、根深蒂固的信念:如果你在这边待久了,那就是一切都闭源、上云,所有流量都会往这个方向走。

And then you spend time in Asia, you know, the the overwhelming belief is that we're going to find the right model to the right workload, and we're not going to overspend.

然后你去亚洲待一阵,那边压倒性的信念是,我们会为对的工作负载找到对的模型,不会超支。

And I think, you know, I would say I would say the next year is probably going to be the most indicative of which way this falls um because I think I think the reason why uh closed source models have captured so much of the value is because um the models actually get the intention and actually carry through the work.

我觉得,明年大概最能说明这事往哪边倒,因为我认为闭源模型之所以拿走这么多价值,是因为这些模型真能领会意图、真能把活儿干完。

And this is the first year where we actually had agents that actually carried out user intention from just answering a chatbot request to actually producing useful work.

而今年是我们第一次真正拥有能执行用户意图的智能体——从只回答一个聊天机器人的请求,到真正产出有用的工作。

Um now the the the level of this intelligent has scaled so rapidly, and we continue to push against like the most economically valuable tasks, which are coding and finance and all these like knowledge work tasks.

现在这种智能的水平扩展得太快了,我们还在不断向最具经济价值的任务推进,也就是写代码、金融,以及所有这些知识工作类的任务。

But like for the long tail of tasks, if open source continues to maintain a 6-month lag, we might actually see a lot more open source used for you know, our everyday tasks that we might actually

但对于任务的长尾,如果开源继续保持六个月的滞后,我们可能真会看到大量开源被用在,比如,我们日常的那些任务上,也许我们真的会——

basically Jensen's argument, right?

这基本上就是 Jensen 的论点,对吧?

Jensen's argument is you're going to have model routing and we're just in a moment in time where the frontier models gain the advantage can do long-running tasks that open source models couldn't do it very well and so they're accruing all of the value, but as soon as the open source models can do the long-running tasks as well, which is not far away that they too will grab a bunch a bunch of this revenue.

Jensen 的论点是,你会有模型路由,我们只是处在一个时间节点上,前沿模型占了优势、能做开源模型做不好的长任务,所以它们拿走了全部价值,但一旦开源模型也能做好长任务——而那一天并不遥远——它们同样会抢走一大块这部分营收。

Are you about to burst into reflection?

你是要突然陷入沉思了吗?

I'm not.

没有。

Okay.

好吧。

No, no, no, no are we, but I'm very impressed by Misha and and the team and what they're doing.

不不不不,不是我们,但 Misha 和他的团队,还有他们在做的事,让我非常佩服。

I very much want a frontier open source US lab to win.

我非常希望一个前沿的、开源的美国实验室能赢。

We know that, you know, I heard you say recently and I believe it to be true Nvidia any day that they really wanted to, right?

我们都知道,我最近听你说过、我也相信这是真的:Nvidia 只要哪天真想做,对吧?

They already have some great open source models.

他们已经有几个很棒的开源模型了。

They could absolutely build a frontier open source model whenever they chose to do it and so it's not a question in my mind as to whether or not the US is going to have a frontier open source model.

他们完全可以在任何他们选定的时候造出一个前沿开源模型,所以在我心里,美国会不会有一个前沿开源模型根本不是问题。

It's just a question about timing and then like at that point in time is that you know, let's say let's assume they get these long-running capabilities.

问题只是时机,然后到那个时间点上,假设他们拿到了这些长任务能力。

Have the frontier labs now achieved something yet again that allows them to keep keep the the stranglehold on the revenues?

前沿实验室是不是又一次做出了什么、让它们能继续死死掐住营收?

Yeah, and I just think it's if you're Wow, that's a cute ASIC you've built there.

是啊,我就觉得,如果你是——哇,你造的这个 ASIC 真可爱。

That is so cute.

太可爱了。

How would you like open source to join the frontier?

你想不想让开源加入前沿?

How would you like that?

你觉得怎么样?

How do you like them apples?

这下你尝到厉害了吧?

So, I mean I'm not sure that's the explicit calculation, but I do think Jensen

我是说,我不确定这是不是那么明确的算计,但我确实觉得 Jensen——

Say more.

展开说说。

Just double click on that for everybody at home.

给屏幕前的各位再细讲一下。

If you were if they were to put an open source model out there, how does that impact the ASIC landscape?

如果你——如果他们真把一个开源模型放出来,那会怎样影响 ASIC 的格局?

Well, you might not have the revenue to fund [laughter] to fund that the revenue of the margins to fund that ASIC.

那你可能就没有营收去资助——(笑)去资助那个 ASIC 了,没有那份营收和利润去养那个 ASIC。

And I do think Nvidia is highly likely to be the world's dominant provider of open source AI.

我确实觉得 Nvidia 极有可能成为全世界开源 AI 的主导供应商。

And I do think Jensen will bring open source, you know, right now it's whatever, 6 months behind the frontier.

我也确实觉得 Jensen 会把开源带上来,现在它大概落后前沿六个月吧。

We might see it creep closer and closer and closer.

我们可能会看到它越靠越近、越靠越近。

And I do think Jensen has a big business decision.

我也确实觉得 Jensen 面临一个重大的商业决策。

I see this, you know, chart here, so let's, you know, chop it up about Nvidia, as you say.

我看到这儿这张图了,那我们就照你说的,好好聊聊 Nvidia。

But if all of his customers are going to compete with him,

但如果他所有的客户都要跟他竞争,

then why not compete with his customers?

那他为什么不去跟他的客户竞争呢?

And we have all these neo clouds.

我们有这么多 neo cloud。

So that's a cloud computing business that can compete with all these cloud computing businesses.

那就是一门云计算生意,可以跟所有这些云计算生意竞争。

He has his own models that are really, really good.

他有自己的模型,而且非常非常好。

Nematron 3 or 3.1 was actually really, really cool from a computer efficiency perspective.

Nemotron 3 或者 3.1,从计算效率的角度看其实非常非常酷。

And he's always careful to release small models so as to not tread on Anthropic and OpenAI,

而且他一向很小心,只发布小模型,以免踩到 Anthropic 和 OpenAI、

Google's toes.

还有 Google 的脚。

But I do think that is a choice he is making.

但我确实觉得那是他在做的一个选择。

And just, you know, at if if the economics change,

而且,你知道,一旦经济账变了,

I think Nvidia can join the frontier and become one of the world's largest cloud computing companies much faster than people think.

我觉得 Nvidia 能加入前沿、成为全世界最大的云计算公司之一,速度会比人们想象的快得多。

Interesting.

有意思。

Interesting.

有意思。

Clark, walk us through this this chart.

Clark,给我们讲讲这张图。

Yeah, so so I think one of the takeaways from spending time in Taiwan was there there is certainly a lot of excitement around the next wave of ASICs.

对,我觉得这趟台湾之行的一个收获是,大家对下一波 ASIC 确实非常兴奋。

Um, but I think I think it's like a very clear moment now where Nvidia it used to be an argument of Nvidia versus ASICs one or the other and, you know, total domination one or the other.

但我觉得现在是一个很清晰的转折点,以前 Nvidia 对 ASIC 的争论是二选一、一方彻底碾压另一方。

Now I think it increasingly every year every every one assumed that Nvidia was going to lose share dramatically on a revenue scale, on a gigawatt scale, on a unit scale.

现在越来越多人每年都假设 Nvidia 会在营收、吉瓦、出货量各个维度上大幅丢失份额。

And actually, if you actually look at the last few years, you know, they've actually maintained their share very, very handsomely.

但实际上,你要是真去看过去这几年,他们其实把份额守得非常非常漂亮。

Um, actually, um, if you accounted for the fact that Anthropic was not really using Nvidia.

而且,如果你把 Anthropic 基本不用 Nvidia 这个因素也算进去。

They probably actually gain share against if not for in 25 26.

在 25、26 年,他们的份额其实是不减反增的。

So, I think I think what was very interesting though was a new class of accelerators or ASICs.

所以我觉得很有意思的是出现了一类新的加速器,或者说 ASIC。

MediaTek with their with their new V8T versus, you know, Broadcom's V8I for TPUs actually was a big topic of discussion.

MediaTek 的新 V8T,对上 Broadcom 给 TPU 做的 V8I,这其实是个讨论得很热的话题。

And, you know, I I think for ASICs the argument now is that more and more will look custom to the actual workload and that is like one vector that people are moving in versus Nvidia now is has kind of shown itself as the the predominant provider of compute to a lot of the world and for, you know, internal internal workloads, perhaps they will go more and more custom and more and more down the stack.

关于 ASIC,现在的观点是越来越多芯片会针对具体的工作负载做定制,这是大家在走的一个方向;而 Nvidia 现在已经把自己坐实成全世界大量算力的头号供应商,但对内部的工作负载,大家可能会越来越走定制、越来越往底层做。

And I I remember just, you know, 1 year ago when it was kind of a Broadcom or Nvidia battle.

我还记得就在一年前,这还是 Broadcom 和 Nvidia 的对决。

It seems there's a lot more nuance now to, you know, what type of accelerators will fit which workloads and fit which customers and fit which business models.

现在感觉里面的门道多了很多:什么类型的加速器适配什么工作负载、适配什么客户、适配什么商业模式。

Um and yeah, I thought I thought that was a a new topic.

嗯,对,我觉得这是个新话题。

It's actually

其实是——

New realization though, I think we all kind of shared this view for a long time.

不过是个新认知,我觉得我们其实很长时间以来都持这个看法。

Yeah, I was just shocked.

对,我当时就是很震惊。

I mean, I'm I'm out here.

我是说,我人就在外面。

I did a board meeting with one of our companies and just, you know, their biggest one thing they emphasized is we thought the world would be have be consuming less Nvidia than it is and if anything, Nvidia is accelerating and they just continue to out execute their competitors.

我参加了我们一家被投公司的董事会,他们最强调的一点就是,我们本以为世界对 Nvidia 的消耗会比现在少,结果 Nvidia 反而在加速,他们就是持续把竞争对手甩在后面。

And I think a lot of people are indexing to this OpenAI gigawatt and you know, Nvidia has 10.

我觉得很多人都盯着 OpenAI 那一吉瓦,而 Nvidia 有 10。

Broadcom has 10. who has six?

Broadcom 有 10。谁有 6?

AMD AMD has six and they have warrants.

AMD,AMD 有 6,而且他们还有认股权证。

And then Cerebras has our shared portfolio company has a gigawatt.

然后 Cerebras,我们共同持有的这家被投公司,有一吉瓦。

And I just that is what's on paper.

我这么说吧,这是纸面上的数字。

What actually gets deployed, let's see.

实际能部署出来多少,走着瞧。

I will be very surprised if you know that 10 out of 27, what's that math?

如果 27 里真只有 10,那我会非常意外,这账怎么算?

Let's see who's best at math.

看看谁数学好。

What percentage market share is that?

那是多少市场份额?

30% yeah.

30%,对。

Yeah.

对。

I'll be very surprised if that is where they land.

如果他们最后真落在那个位置,我会非常意外。

I think that is an extremely unlikely outcome.

我觉得那是个极其不可能的结果。

And especially as long as we're in a watt constrained world, if you can get more tokens per watt, which is literally revenue with Nvidia than a lot of alternatives just if you build your factory with another chip you may save some money, but you're going to have less revenue and the margins may be lower and that's a point that Jensen keeps hammering and I think is a really important.

尤其是只要我们还处在受电力约束的世界里,如果你能榨出更多每瓦 token 数——用 Nvidia 这就等于营收——比很多替代方案都高;你要是用别家芯片建你的工厂,可能省了点钱,但营收会更少,毛利率可能也更低,这正是 Jensen 一直反复敲打的一点,我觉得非常重要。

And by the way, credit where credit is due the most important the most one of the most surprising things to me in this ASIC landscape

顺便说一句,该给的功劳得给,在这个 ASIC 格局里,最让我意外的事情之一——

I'd say Meta and Microsoft have been probably disappointing.

我得说 Meta 和 Microsoft 大概是让人失望的。

You know who made a good ASIC?

你知道谁做出了一颗好 ASIC 吗?

Well, I know you know.

我知道你知道。

Jalapeno

Jalapeno。

Yeah, exactly.

对,正是。

from Open AI.

来自 OpenAI。

They made a great chip.

他们做出了一颗很棒的芯片。

Now, unfortunately needs to run at a much lower temperature than the Nvidia GPUs, which means you need to spend more money on cooling and that consumes more power.

只不过,可惜它得跑在比 Nvidia GPU 低得多的温度下,这意味着你得在散热上多花钱,而散热又更耗电。

They made a great chip.

他们做出了一颗很棒的芯片。

Well, we can I mean I think the question there and the question for everybody is going to be is that the highest and best use of your time?

嗯,我们可以——我觉得这里的问题、也是所有人都要面对的问题是:这是不是你时间的最高、最优用途?

Right?

对吧?

Like I you know, I tend to think that the frontier companies like there's this belief that they got to be vertical vertically integrated.

我倾向于认为,那些前沿公司——有一种观点认为它们必须垂直整合。

But if you believe like I do that the race to super intelligence particularly as we get these recursive loops working may be over in the next two to three years, then I think focus focus focus focus.

但如果你像我一样相信,通往超级智能的竞赛——尤其是当这些递归循环开始跑通——可能在未来两三年内就结束,那我觉得就该聚焦、聚焦、再聚焦。

You exist to build the best intelligence in the world and to deliver the best intelligence in the world and you that means you have to have all the revenue.

你存在的意义就是造出全世界最好的智能、交付全世界最好的智能,而这意味着你必须拿下所有营收。

Because if you want to build out the compute that's going to be required to continue to push the frontier, you have to have the revenue in order to support it.

因为如果你想建起继续推动前沿所需的算力,你就必须有营收来支撑它。

So I think you know, subject to the focus question, I think they certainly did.

所以我觉得,在聚焦这个问题的前提下,他们确实做到了。

This all brings me back to kind of a reality check, though. you know, we just got done talking about test time compute, inference time compute, long-running agents.

不过这一切又把我拉回到一个现实检验。我们刚聊完测试时算力、推理时算力、长时运行的智能体。

This is really the thing that's unlocked the revenue this year.

这才是今年真正解锁营收的东西。

Um it all pushes us in the direction of more CapEx.

而这一切都把我们推向更多的 CapEx。

Google just raised $80 billion, right?

Google 刚融了 800 亿美元,对吧?

We've now taken the Mag 5 or Mag 7 free cash flow, you know, down dramatically, 80% um from just a few years ago.

我们现在把 Mag 5 或者说 Mag 7 的自由现金流,大幅拉低了,比几年前少了 80%。

Um and Morgan Stanley, you've got this chart in front of you, up to their 2027 CapEx forecast from 950 billion to 1.1 trillion.

还有 Morgan Stanley,你面前就摆着这张图,他们把 2027 年的 CapEx 预测从 9500 亿上调到了 1.1 万亿。

I mean, we were talking about this with Jensen.

我们跟 Jensen 也聊过这个。

That was his forecast 2 years ago.

那是他两年前的预测。

You know, obviously, this doesn't even include SpaceX, CoreWeave, etc.

而且很明显,这甚至还没算上 SpaceX、CoreWeave 这些。

So, I think the number on 2027 is likely closer to 1.5 trillion.

所以我觉得 2027 年的数字很可能更接近 1.5 万亿。

And if we compare this to the total incremental inference revenue, so the thing that the market gets worried about, you know, back to my Sam Altman podcast, you know, in October of last year, can we really afford to spend 1.5 trillion of CapEx a year if we're only generating X amount in inference revenue?

如果把这个跟总的增量推理营收对比——也就是市场担心的那个点——回到我去年十月那期跟 Sam Altman 的播客,如果我们一年只产生 X 那么多推理营收,真的花得起一年 1.5 万亿的 CapEx 吗?

The thing I think that lit the fuse this year was Anthropic showed up in a major way with revenue, right?

我觉得今年点燃引线的,是 Anthropic 带着营收强势登场,对吧?

And so, we have, you know, the AI lab revenue everybody combined at around $300 billion next year, right?

所以我们现在,把所有 AI lab 明年的营收加起来大概是 3000 亿美元,对吧?

So, can't you know, and go roll that out to 2027 uh or that is 2027, 300 billion.

那——把这个推到 2027 年,或者说那就是 2027 年,3000 亿。

So, we're spending 1.5 trillion of CapEx on 300 billion of inference revenue.

所以我们是在用 1.5 万亿的 CapEx,去对 3000 亿的推理营收。

Does that math math for you?

这账你算得平吗?

And what would cause you, you know, to to get more nervous again about our ability to continue to make these investments?

什么情况会让你重新对我们持续做这些投资的能力更紧张起来?

Because the second we get nervous about it, the entire semi complex is going to come down a lot.

因为我们一旦开始紧张,整个半导体板块就会大幅下挫。

Well, what do you think the gross margins are on that 300 billion?

那你觉得那 3000 亿的毛利率是多少?

Yeah, let's call it 50%.

就算它 50% 吧。

I I would guess they're probably a little bit higher than that.

我猜可能还比这个高一点。

I might say 60 or 70.

我会说 60% 或 70%。

But, I mean, that math starts to math, and what I would just say is I think that 300 billion is low, man.

但这样一算,账就开始算得平了,而我想说的是,我觉得 3000 亿这个数偏低了,老兄。

Yeah.

是啊。

Yeah.

对。

I just think it's low.

我就是觉得它偏低。

From your mouth to God's

但愿如你所说——

Yeah, yeah, exactly.

对对,正是。

I think I think we end this year well over 200 billion in inference revenue, well over.

我觉得我们今年结束时推理营收会远超 2000 亿,远超。

And so, I think the math really maths, and I do think we have to give uh Jensen

所以我觉得这账真的算得平,而且我觉得我们得给 Jensen——

Yeah, our friend.

对,咱们的朋友。

some credit because he said some things that seemed outlandish.

——一点认可,因为他说过一些当时听起来很离谱的话。

And he was conservative.

而他其实是保守的。

He was low.

他说低了。

He said a trillion 2 years ago.

他两年前说的是一万亿。

And I mean, he was really low.

说真的,他说得太低了。

And so, like, let's give the guy some credit and think about what he is saying right now.

所以,咱们该给他一点认可,好好想想他现在正在说的话。

For sure, for sure.

绝对,绝对。

And and listen, I would say consistently, Elon's been taking the over.

而且听着,我得说一直以来,Elon 都在往高了押。

Sundar's been taking the over.

Sundar 也在往高了押。

Sam, Dario, you know, Dario did the podcast with Dwarkesh when he was talking about country geniuses in the data center.

Sam、Dario——你知道 Dario 上过 Dwarkesh 那期播客,他聊到数据中心里的天才之国。

He said that will be here by 2028.

他说那会在 2028 年到来。

He said revenues will go into the low hundreds of billions by 2028.

他说到 2028 年营收会进入千亿出头的量级。

So, let's call that, you know, 3 400 billion of revenue by 2028.

所以就算它到 2028 年是三四千亿的营收吧。

And he said that a while ago now, so he may even be revising up his number.

而且他这话是有一阵子之前说的,所以他甚至可能还在上调他的数字。

And he said it's hard for me to see that there won't be trillions of dollars in revenue before 2030.

他还说,我很难想象 2030 年之前不会有数万亿美元的营收。

And if you're on that revenue trajectory, if we're on a trajectory to 200 by the end of this year, let's call it 4 or 500 by next year, and a path to trillion plus by 2029, then the math maths.

如果你在那条营收轨迹上——如果我们年底奔着 2000 亿去,明年就算四五千亿,再到 2029 年有一条通往一万亿以上的路径,那这账就算得平。

And we got to keep in mind that half of the spending is there to you know, for training, maybe a little less than half.

而且我们得记住,其中一半的支出是用来训练的,也许不到一半。

What is it, Foxy?

是多少来着,Foxy?

It's probably that depends on the lab, but it's I would say it's increasingly less than half.

这大概——要看是哪个 lab,但我会说越来越低于一半了。

Okay.

好。

So, we'll call it 35% is spending that's not revenue generating, but it's going to kind of make the next model.

那我们就算 35% 是不产生营收的支出,但它是用来造出下一代模型的。

So, I think the math maths.

所以我觉得这账算得平。

And there's still this prisoner's dilemma where if you opted out, that may be an existential decision.

而且这里还存在一个囚徒困境:如果你选择退出,那可能是个关乎生死存亡的决定。

And I think like coming into this year, going back to this kind of what narratives were violated, you know, I think into this year everyone expected token pricing, uh the price of compute, it's all deflationary.

我觉得,进入今年的时候,回到刚才说的哪些叙事被打破了——我觉得进入今年时,所有人都以为 token 定价、算力价格,都是通缩的。

And it will be kind of a smooth line deflationary over time.

而且会是一条随时间平滑向下的通缩曲线。

But, I think this year what we've seen is the opposite.

但我觉得今年我们看到的恰恰相反。

And you know, it's all comes back to supply-demand.

这一切归根结底还是回到供需。

The demand side of the equation seems to be far outstripping the supply.

这个等式里的需求端,似乎远远超过了供给。

Right?

对吧?

And I think you look at the deals signed by SpaceX and others, the monetization rates per watt are increasing. and look, that is on a a pretty nascent small base of users, right?

我觉得你看 SpaceX 和其他家签的那些单子,每瓦的变现速率在上升。而且注意,这还是建立在一个相当初期、很小的用户基数上的,对吧?

Like Alex at Well Rock, he has this great um way to frame it.

比如 Wellrock 的 Alex,他有个特别好的框架来描述这件事。

Less than 0.2% of people on Earth are actually using AI in an agentic way.

地球上真正以智能体方式使用 AI 的人还不到 0.2%。

Right?

对吧?

Like I'm not a technical person, but I'm consuming 500 CPU cores in a VM instance, five GPUs 24/7.

比如我不是技术人员,但我一个 VM 实例里就在消耗 500 个 CPU 核心、五张 GPU、7×24 小时不停。

I mean, if you draw that out to any meaningful percentage of the population, I mean, we're going to be in, you know, this kind of shortage environment maybe for some time.

如果你把这个推演到人口里任何有意义的比例,那我们可能会在这种短缺的环境里待上一段时间。

So, I think that is all positive for this ROI question.

所以我觉得这对 ROI 这个问题都是正面的。

Man, foxy, 100 to one CPU to GPU ratio.

老天,Foxy,100 比 1 的 CPU 对 GPU 比例。

Kind of agentic workflow.

属于智能体式的工作流。

He said of course.

他说当然。

Five.

五。

Five, yes.

五,对。

I'm being smart with my phone.

我在很聪明地用我的手机呢。

Good, good, good.

好,好,好。

Excellent.

太棒了。

I I will say also that ratio of 300 to 1.

我还得说一句,那个 300 比 1 的比例。

You know, call it 1.2, 1.5.

就算它 1.2、1.5 吧。

Um there there is also a rate that now physically we can only expand how much we can produce and how much we can actually increase that spend by, whereas we're seeing the opposite right now on the on the the willingness to pay for these tokens.

而且现在还有一个速率的问题:物理上我们能扩产多少、能实际把支出增加多少,是有上限的;而我们现在在为这些 token 付费的意愿上看到的却是相反的走势。

And actually like when the willingness to pay for these when the monetization per gigawatt is actually increasing from, you know, call it like 20 20 billion um in the in the best best of cases for at the beginning of the year to now like 30 to even pushing 40

而实际上,当为这些付费的意愿——当每吉瓦的变现从年初最好情况下大概 200 亿、200 亿,涨到现在的 30 亿甚至逼近 40 亿——

per gigawatt

——每吉瓦——

per gigawatt.

——每吉瓦。

Um all of that is is a very heavy fixed cost base, but all of that is like pure margin flow through now.

这一切都建立在非常沉重的固定成本基础上,但这一切现在都是纯利润的流入。

And you're actually, you know, as we scale like the willingness to pay for for all of this and and now the all of this stipulated by like, you know, everything we're talking about of like how much is open source versus not and all of these different flows, but really like as we're climbing this curve, you know, the the the revenue is might actually outstrip our fixed cost base by by significant amount.

而实际上,随着我们扩大规模,为这一切付费的意愿——现在这一切都取决于我们聊的那些,比如多少是开源、多少不是,以及各种不同的流向——但真的,随着我们爬上这条曲线,营收其实可能会大幅超过我们的固定成本基础。

And I think that's why all the labs are pushing, you know, the the the gas to the pedals because they all they all see like within if we continue this curve within like 3 years, you know, we're just going to be so short on all the computer

我觉得这就是为什么所有 lab 都在把油门踩到底,因为它们都看到——如果我们沿着这条曲线走下去,大概三年内,我们的算力就会极度短缺——

It's a great I'm sorry, but I mean I like it's a great point.

这是个很——抱歉,我是说,我觉得这是个很好的点。

Like if you thought you were getting a when you made these decisions

就是说,如果你当初做这些决定的时候以为你能拿到——

in November of 2025, you thought you were getting a certain return.

——在 2025 年 11 月,你以为你能拿到某个特定的回报。

You may be getting triple that return today.

而今天你拿到的回报可能是那个的三倍。

At Tropic, no way no way did they think they were going to be anywhere close to break even.

Anthropic 当初绝无可能——他们绝没想到自己会接近盈亏平衡。

Right?

对吧?

And and and and in this part of the curve, and the reason like I I I called it accidental profitability that, you know, people have been talking about that because they want to spend a lot more money on computer.

而且在曲线的这个阶段,我之所以把它叫作意外盈利——大家一直在聊这个,是因为他们其实想在算力上花多得多的钱。

They just had a hard time doing it.

他们只是很难花得出去。

Now maybe with SpaceX, you know, they could take some of those dollars and and and go spend them other places.

现在也许有了 SpaceX,他们可以把其中一部分钱拿去别的地方花。

But that to me is, you know, a a fundamental change.

但在我看来,这是个根本性的转变。

Um the first argument against the frontier labs was they'll never generate revenue.

反对前沿 lab 的第一个论点是,它们永远产生不了营收。

Okay?

对吧?

And then we that got blown up.

然后这个论点被打爆了。

Then it was like even if they generate revenue, it'll be really shitty gross margins, and they'll never be able to get make money.

接着就变成:就算它们能产生营收,毛利率也会烂得要命,永远赚不到钱。

And then kind of that that's blown up.

然后这个论点也差不多被打爆了。

And and you know, I think now, you know, people are falling back and they're saying, "Well, they're overcharging.

于是我觉得现在,大家又退回来说:“它们收费太高了。

This is token maxi." My good friend, you know, Chamath has said there's no ROI on any of this spend.

这就是 token maxi。”我的好朋友 Chamath 就说过,这些支出没有任何 ROI。

It's all this token maxi.

全都是这种 token maxi。

My best evidence for why we all know, of course, when somebody puts on this much spend like at Altimeter, we're not optimally spending every single dollar.

而我能拿出的最好证据是——我们当然都知道,当有人像 Altimeter 这样投入这么多支出时,我们并不是每一块钱都花在最优处。

But, the question is, why are millions of independent businesses, small, medium, and large, why are millions of consumers all choosing to do the same thing?

但问题是,为什么数以百万计各自独立的企业,小的、中的、大的,为什么数以百万计的消费者,都同时选择做同一件事?

They're not dumb.

他们不傻。

These are, you know, rational economic actors that are all simultaneously saying, "I want to do this because it makes my life better.

这些都是理性的经济行为体,他们都在同时说:“我想做这件事,因为它让我的生活更好。

It makes my business better, etc." To me, that is the best evidence as to why I think this revenue can continue.

它让我的生意更好,等等。”在我看来,这就是为什么我认为这份营收能持续下去的最好证据。

Yeah.

对。

And Clark, I think like the point you made is dead on cuz I mean, you want to own asset-heavy businesses in inflationary environments, and token pricing is going up, and supply and demand is tightening, so totally agree.

Clark,我觉得你刚才那个点说得太准了,因为在通胀环境里你就是想持有重资产的生意,而 token 定价在涨,供需在收紧,所以完全同意。

Um you know, as we begin to uh find our way to the exit ramp and and [laughter] and wrap here, one of the things I you know, you and I've been doing this for a long time, Gavin, a couple decades.

那我们开始找出口、准备收尾了。有件事我想说,Gavin,你我干这行都很久了,几十年了。

Um you may even sketch longer than me, even though I'm a little bit older than you.

你可能比我干得还久,虽然我年纪比你大一点。

Um you know, we have uh I always like to do a market check, because I find a lot of time that analysts come on these things, and they talk their, you know, talk their book, and you know, there are a lot of people who listen to these things, retail investors and others.

我一直喜欢做个市场盘点,因为我发现很多时候分析师上这类节目,只说对自己持仓有利的话,而听这类节目的人很多,散户之类的。

It's just kind of like, what do we really think?

就是想问,我们到底怎么想?

And so, I always characterize as kind of small, medium, and large.

我一般会把仓位分成小、中、大三档。

Like, what am I doing?

就是我现在在干嘛?

Do I have small exposure on?

我是小仓位吗?

Do I have medium exposure on?

我是中等仓位吗?

Do I have large exposure on?

我是重仓吗?

You know, and if you look at what's happened in the markets, semis ripped this year.

你看今年市场发生了什么,半导体今年暴涨。

I mean, like uh you've been doing this a long time.

你也干这行很久了。

I don't I've never seen it before, right?

我以前从没见过,对吧?

I've never seen, you know, the doubles and the triples across the board like we saw.

我从没见过像这样全线翻倍、翻三倍的行情。

But, there's been huge dispersion, right, in the market.

但市场里出现了巨大的分化,对吧。

Internet's down 16%, uh software's down 8% on the year.

互联网板块今年跌了 16%,软件跌了 8%。

You know, spy and and Nasdaq are up, but really up because of their components that are related to AI and compute.

标普和纳斯达克是涨的,但真正拉涨的是里面跟 AI 和算力相关的成分股。

And so, the market itself has kind of struggled.

所以市场本身其实挺挣扎的。

Meanwhile, if you were in the stuff that we were invested in, we've all done pretty well.

与此同时,如果你持有的是我们投的那些东西,大家都做得不错。

I think you know, I've said it a couple times.

我说过好几次了。

I think if the Anthropic revenue had not shown up this year, because that was the overhang on the market, I think the whole market could be down this year.

我觉得如果今年 Anthropic 的营收没出来——因为那是压在市场头上的一块石头——整个市场今年可能是跌的。

Right?

对吧?

Um but that showed up.

但它出来了。

You know, we just had these huge months in in in April and May.

我们四五月份就迎来了这几个大涨的月份。

Um for us, you know, because prices came up so much, because I have some worry about, you know, geopolitics, the macro backdrop with, you know, with with what's going on with inflation in the short run, and just like, you know, needing a little consolidation in this market to answer some of these questions, because now expectations are higher.

对我们来说,因为价格涨了这么多,因为我对地缘政治有点担心,对短期通胀这个宏观背景有点担心,也因为现在预期更高了,市场需要一点消化来回答这些问题。

You know, we dialed back from what I would call large for Altimeter to something kind of like medium small.

我们从我称之为 Altimeter 的重仓,回调到了偏中小的仓位。

Um again, it's never all or nothing for us.

再说一次,对我们来说从来不是全有或全无。

It's like, what is the risk-reward at a given price?

而是在某个价格下,风险回报比是多少?

Um and so, we think this is a, you know, maybe going to be a period of consolidation on way to much higher highs.

所以我们觉得这可能是一段消化期,通往更高的高点。

Um curious just how you run the book, how you think about it like a portfolio manager.

我很好奇你怎么打理仓位,作为组合经理你怎么看。

Very similarly, man.

非常相似,老兄。

I always think stocks, the markets, I imagine them as runners.

我一直把股票、市场想象成跑步的人。

Okay?

好吧?

And like in '22, that runner had gone downhill.

比如 22 年,那个跑步的人在下坡。

It had a lot of energy, man.

他积攒了很多能量,老兄。

Yeah, it was painful.

对,那很痛苦。

It wasn't fun.

一点都不好玩。

Um but coming out of that, there was a lot of kind of pent-up upside in the market.

但走出来之后,市场里积压了很多向上的动能。

And you know, the market, particularly last 2 months, it has run up a very steep hill.

而市场,尤其是最近这两个月,一路冲上了一个很陡的坡。

And a lot of companies, semiconductor companies in particular, you know, ironically, you know, Nvidia and Broadcom, they they have been laggards.

很多公司,尤其是半导体公司,讽刺的是,Nvidia 和 Broadcom 反而是落后的。

And so, but a lot of these, like I do see a lot on X about finding the next bottleneck.

但很多这类的,我在 X 上看到很多人在找下一个瓶颈。

I think that was the last game.

我觉得那是上一局的玩法了。

That game is over.

那一局已经结束了。

You've had a lot of stocks that forget climbing a mountain or a hill.

有很多股票别说爬山爬坡了。

They've gone straight up a cliff, okay?

它们是直接冲上了一面悬崖,好吧?

Yes.

是的。

They're tired.

它们累了。

They need to rest.

它们需要休息。

And we'll see, do they just rest at the top of that cliff they climbed?

我们等着看,它们是就在爬上去的那面悬崖顶上歇着?

Do they hang out on the in their harness for a while?

还是挂在自己的 harness 上待一阵?

We've seen some.

我们见过一些。

Or do they need to go downhill for a bit?

还是需要下坡走一段?

We'll see, but I'm thinking very similarly to you.

我们走着瞧,但我的想法跟你非常一致。

But it is and I think there's, you know, the market is seasonal.

但确实,而且我觉得市场是有季节性的。

I think there's real real concerns around inflation and rates.

我觉得围绕通胀和利率有实实在在的担忧。

What was CPI this morning?

今早的 CPI 是多少来着?

uh 4.2.

4.2。

I think we added core came in at like 0.2 versus 0.3, so a little bit better.

我记得核心 CPI 是 0.2,对比之前的 0.3,所以好了一点。

Um but you know, clearly we're we're above four again.

但显然我们又回到了 4 以上。

And um and and there's short-term pressure on, you know, core PCE, etc.

而且核心 PCE 之类的短期有压力。

Um and we have some unknown unknowns, but the market, I mean, if I had told you the fact pattern for this year, that we're going to be in a war with Iran, that, you know, oil was going to be at 100 bucks, that CPI was going to be creeping back up, that internet was going to be down 15%.

我们还有一些未知的未知,但这个市场——我是说,如果我早跟你说今年会是这样一套剧情:我们会跟伊朗打仗、油价会到 100 块、CPI 会重新往上爬、互联网板块会跌 15%。

Software is going to be down 8%.

软件会跌 8%。

You would have said, "I want nothing to do with that market, right?" And here we are.

你会说:"这种市场我碰都不想碰,对吧?"结果我们现在就在这儿。

The market's done pretty good in the stuff that we traffic in because the world underestimated AI revenues and underestimated the amount of compute that was going to be needed.

市场在我们涉足的东西上表现相当好,因为全世界低估了 AI 营收,也低估了会需要的算力规模。

It's odd to say you know, we're heading into a seasonally weak period with all of these fears.

说来奇怪,我们正带着这一堆恐惧走进一个季节性偏弱的时段。

AI has actually been seasonal for the last three summers.

AI 过去三个夏天其实都有季节性。

Token consumption is kind of plateaued, slowed down, and that's cuz you know, college kids are big AI consumers and they don't use as much AI, you know, hopefully they're all using it to learn and not cheat.

token 消耗有点见顶、放缓了,那是因为大学生是 AI 的消费大户,放假就不怎么用 AI 了,但愿他们都是用来学习而不是作弊。

But that may happen.

不过那也可能发生。

It may not happen because of generative AI.

也可能因为生成式 AI 而不会发生。

is building swarms of agents, building a SpaceX model.

在构建智能体集群,构建一个 SpaceX 模型。

He's going to the SpaceX IPO with me at the exchange on Friday, but I he had to build an AI model using AI agents.

周五他要跟我一起去交易所参加 SpaceX 的 IPO,但他得用 AI 智能体做一个 AI 模型。

He had to build a model, a DCF before we go to the exchange.

他得在我们去交易所之前做一个模型,一个 DCF。

He is mesmerized.

他着了迷。

He is absolutely and it's extraordinary what he's doing.

他完全,他做的东西太不可思议了。

So he's one kid who's not easy to less computer

所以他算是一个不容易少用电脑的孩子

or something.

之类的。

He's burning it.

他在猛烧。

He's burning it.

他在猛烧。

Yeah, but you know, if token consumption plateaus, if open source takes some share, there's a Silicon data index that has showed, which is an index of kind of consumption and pricing.

对,但你知道,如果 token 消耗见顶,如果开源抢走一些份额——有个 Silicon Data 指数显示,它是个反映消耗和定价的指数。

I think there may have been a little bit of a shift over the last 2 weeks to open source tokens that are cheaper.

我觉得过去两周可能出现了一点向更便宜的开源 token 的转移。

Like people looking at that data as bearish or not understanding it.

就是有人把那个数据看成看空信号,或者没看懂。

But nonetheless, like I just think there's reasons, you know, to look around, be careful, be thoughtful.

但不管怎样,我就是觉得有理由四处张望、小心一点、想周全一点。

I always assume a bullet is coming for me.

我总是假设有一颗子弹正朝我飞来。

Head on [laughter] a swivel.

时刻警惕、四处张望。

It's the bullet you don't see that gets you.

打中你的往往是你没看见的那颗子弹。

So, I'm trying to spin as fast as I can.

所以我在尽量快地旋转躲闪。

But yeah, it's the market may need to take a breather.

但没错,市场可能需要喘口气。

But man, when I think about what Noam Brown said and when I see the capabilities of Fable, it's just hard for me to get too bearish.

但老兄,当我想到 Noam Brown 说的话,当我看到 Fable 5 的能力,我实在很难变得太看空。

I mean, like to me um and we got two, I think, of the most extraordinary guys of, you know, the next generation, you know, sitting in the room.

对我来说,我们这屋里坐着我觉得是下一代最了不起的两个人。

We have at Altimeter, we have deep admiration for the work that you guys do.

我们 Altimeter 对你们做的工作深深敬佩。

I always appreciate when you send me a note about the work that we do and we publish.

每次你给我发条便条评论我们发布的工作,我都很感激。

Um but for the guys who are newer to the business, they might think this is the way that it kind of always was, right?

但对那些入行更晚的人来说,他们可能会以为一直以来都是这样的,对吧?

And like this line, the steepening of the line of creative destruction, the steepening of the line of, you know, scale advantages.

而这条线,创造性破坏这条线的陡峭化,规模优势这条线的陡峭化。

Um I always believed it was to it was going to be true.

我一直相信它会成真。

I never thought it would be true at this rate.

但我从没想过会以这样的速度成真。

I went back last night.

我昨晚回头查了下。

In the last 7 years, we've added 1 trillion of revenue to the Mag 7 in the last 7 years, okay?

过去 7 年,我们给 Mag 7 增加了 1 万亿美元营收,就在过去这 7 年,好吧?

To get to a trillion, to get to the first trillion of, you know, took over 20 years.

而要做到 1 万亿,做到第一个 1 万亿,花了 20 多年。

In the last 7, we had another tr- trillion and that added 17 trillion in market cap.

过去 7 年里我们又加了一个万亿,而这带来了 17 万亿的市值。

That trillion dollars, okay?

就那 1 万亿美元,好吧?

I The forecast now that we're going to add another trillion of revenue in just three companies SpaceX Anthropic and open AI over the next four to five years.

现在有预测说,未来四到五年,光是 SpaceX、Anthropic 和 OpenAI 这三家公司,就会再增加一个万亿的营收。

Okay, like not seven companies three companies and in half the time right and so I would say that you know, we are going to have bumps in the road.

好吧,不是七家公司,是三家公司,而且时间只有一半,对吧,所以我会说,路上肯定会有颠簸。

I know that it's going to be like this but we're going to higher highs because the size of the prize.

我知道过程会是这样,但我们会走向更高的高点,因为这个奖赏太大了。

This is going to transform five ten 15% of global GDP.

这将改变全球 GDP 的 5%、10%、15%。

There is no doubt in my mind and 10% of global GDP is 10 trillion dollars.

我心里毫不怀疑,而全球 GDP 的 10% 就是 10 万亿美元。

It's an exciting future to be a part of it's fun to do it with you guys.

能参与其中是个激动人心的未来,和你们一起做这件事很有意思。

I think we're going to have to do our work to do the things to make sure America wins and that we evolve the social contract keep everybody you know lift the floor take everybody with us on this ride but it's a it's a it's a really exciting time to be doing what we're doing it's fun to be doing it with you guys.

我觉得我们得做好自己的功课,做该做的事,确保美国赢,确保我们让社会契约进化,把所有人都带上,抬高底线,带着每个人一起上这趟车,但这真是一个非常激动人心的时刻做我们在做的事,和你们一起做很有意思。

Yeah, I just want to say Brad thanks for having us and thank you for what you've done with the Trump accounts.

对,我就想说,Brad 谢谢你请我们来,也谢谢你为 Trump accounts 做的事。

I actually think it's super important for America for the world to give people an equity stake at a very young age.

我其实觉得,在很小的年纪就给人一份股权,对美国、对世界都超级重要。

They they will see it compound over their lifetimes.

他们会看着它在一生中不断复利增长。

This is a great thing you've done for the world.

这是你为世界做的一件了不起的事。

So thank you.

所以谢谢你。

I'd echo all your comments like deep admiration for you your team gratitude for the collegiality and friendship between our firms.

我完全呼应你的话,对你和你的团队深深敬佩,也感激我们两家公司之间的同行情谊和友谊。

I know Clark and Foxy they hang out like all the time.

我知道 Clark 和 Foxy 他们几乎天天混在一起。

That's a people think that you know and there are people in our business who don't want to share anything.

这一点,大家以为——我们这行有些人什么都不愿意分享。

Our view is like we open source it but there are very few people who we actually call and ask their opinion because there are very few people who do the thousands of hours of work that we do you know that are adding to that and you do it and we appreciate that and you do as well Gavin we appreciate that.

我们的看法是开源分享,但真正会打电话去问意见的人非常少,因为真正投入我们这样几千小时工作、给这件事添砖加瓦的人非常少,而你们做到了,我们很感激,你也一样,Gavin,我们很感激。

So with that love fest let's call it a wrap.

那就带着这份互相吹捧,我们收尾吧。

Thanks for being here.

谢谢你们来。

Thank you.

谢谢。